Sei sulla pagina 1di 400

E Q U I T Y

R E S E A R C H :

U N I T E D S T A T E S

Data Storage Infrastructure


October 22, 2001

United States
H. Clinton Vaughan
415-951-1786
h.clinton.vaughan@ssmb.com The SAN Book 3.0
John C. Dean
415-951-1622
(Storage Area Networks)
john.c.dean@ssmb.com
Everything You Ever Wanted to Know About
Joseph Wana
415-951-1804
Storage (but Couldn’t Find the Answer)

Scott McBride ➤ This book examines the intricacies and


415-951-1829
evolutions of today’s data storage market that
Michael Cadiz investors need to know, including storage
415-951-1671
subsystem technologies, file system technology,
virtualization, emerging software “killer
applications,” switching architectures, iSCSI, and
InfiniBand.

➤ We also detail a bunch of large corporations’ IT


infrastructures to provide real world storage
experiences, trends, and expectations.

➤ Storage networking is here! The debate has


moved from if it’s going to happen, to how.

➤ We expect storage networking and software to


lead and drive the broader storage industry’s
growth.

➤ We believe that storage will be an outperforming


industry.
The SAN Book III – October 22, 2001

COMPUTER STORAGE ARCHITECTURES


Traditional Architecture = Client: PC and UNIX workstations

= Server: Mainframe, UNIX and NT

= Disk Subsystems: RAID and JBOD


LAN
= Tape Subsystems: Drives and Libraries

NAS = Network Attached Storage


SAN = Storage Area Network

NAS Architecture SAN Architecture

LAN LAN SAN

NAS

Source: Salomon Smith Barney

2
The SAN Book III – October 22, 2001

Bottom line: Data continue to grow. We believe that this


fundamental principle will continue to drive storage industry
growth over the long term. In addition, we expect to continue
seeing applications — such as B2B, CRM, ERP, and supply chain
software — become more data intensive. This growth in data will
require greater management and newer technologies; thereby,
offering opportunity to companies that can add value through
innovation.

We believe SANs will drive storage growth.


Just as Local Area Networking (LAN) drove growth in the 1990s for PCs and
servers and increased the use of computing environments, driving significant
infrastructure investments in the process, we believe Storage Area Networking
(SANs) will increase the use of data and drive storage growth over the next decade.
We believe that storage networking and software will lead and
drive the industry’s growth.
We believe the storage networking market is less than 10% penetrated and will
offer excellent growth in the years to come. In our opinion, the storage
software market is on the cusp of delivering the killer applications that will
unlock many of the benefits of networking storage. Both of these trends
combined will drive the adoption of newer technologies and enable storage
suppliers to create and deliver added value to customers for many years (i.e.,
storage is not a commodity).
We expect to see an abundance of new technologies sweep the
market.
We expect to see leading players continue to gain momentum due to the
increasing complexity of storage technology. This should bode well for
industry leaders EMC, Veritas, and Network Appliance. This should also give
legs to Fibre Channel, although IP storage is clearly on the way. We also
expect startups to help shape the storage industry of the future due to their
intense focus on emerging technologies.
Replication should play an increasing role.
While the past year has been focused on increasing capacity utilization rates
and lowering overall storage cost, we expect the upcoming years (2002 and
beyond) to focus on replication which, in effect, will double the demand for
data by requiring multiple copies.
Reading tips for this report.
We have designed this report to be an easy read for beginners and an intriguing
one for experts. Even though this is our third edition, we truly believe there is
something to be learned in every section of this book for every level of reader.
Special thanks to Heather King, our summer 2001 intern, who made huge
contributions to this report.

3
The SAN Book III – October 22, 2001

Table of Contents
Executive Summary .................................................................................................................................... 3
DEMAND DRIVERS................................................................................................................................ 5
State of the Storage Market ....................................................................................................................... 6
Demand Drivers: The Real World ......................................................................................................... 14
Demand Driver Analysis .......................................................................................................................... 30
Data Sheet.................................................................................................................................................. 45
SAN AND NAS..................................................................................................................................... 47
What Is a SAN?......................................................................................................................................... 48
What Is NAS?............................................................................................................................................ 62
SAN and NAS Convergence ..................................................................................................................... 67
What Is Server Clustering?...................................................................................................................... 72
SUBSYSTEMS ..................................................................................................................................... 77
Subsystems: The Heart of Storage ......................................................................................................... 78
Disk Drives to Subsystems........................................................................................................................ 82
RAID and Beyond..................................................................................................................................... 90
NAS Subsystems...................................................................................................................................... 104
RAID Is No Good Without Core Software ........................................................................................... 112
SOFTWARE ....................................................................................................................................... 133
The Magic Word Is Software.................................................................................................................. 134
Software Applications............................................................................................................................. 147
NETWORKING................................................................................................................................... 171
Storage Networking ................................................................................................................................ 172
Storage Networking I/O: HBAs and ICs............................................................................................... 178
Storage Networking Equipment ............................................................................................................ 194
PROTOCOLS ..................................................................................................................................... 225
Protocols: The Language of SANs........................................................................................................ 226
The Basics of I/O Interfaces ................................................................................................................... 231
Fibre Channel: Networking Storage .................................................................................................... 236
Fibre Channel Versus Ethernet ............................................................................................................. 239
Potential Future IP Storage Iterations.................................................................................................. 249
The Future Band: InfiniBand ............................................................................................................... 255
Recent Company Developments ............................................................................................................ 266
COVERAGE UNIVERSE..................................................................................................................... 273
Companies in Our Universe ................................................................................................................... 274
Coverage Universe .................................................................................................................................. 275
PRIVATE COMPANIES ...................................................................................................................... 281
GLOSSARY........................................................................................................................................ 367
TABLE OF FIGURES.......................................................................................................................... 391

4
The SAN Book III – October 22, 2001

Demand Drivers

State of the Storage Market


Demand Drivers: The Real World
Demand Driver Analysis
Data Sheet

5
The SAN Book III – October 22, 2001

State of the Storage Market


➤ We continue to view storage as a great place to invest. We believe it
is an excellent infrastructure play and expect it to be an outperformer
well into the future.

➤ We expect to see storage networking and software lead the


industry’s overall growth.

Information is like corporate


The Storage Market in 2001 and Beyond
oxygen — the life-blood a Even in a global economic slowdown, information is still an organization’s most
company needs to operate valued and critical asset. We think of it as the oxygen, or the life-blood a company
efficiently and make sound needs to survive and make sound business decisions in an increasingly competitive
business decisions in an
increasingly competitive
environment. Today in a typical organization, information comes from many
environment. sources: Internet applications, e-mail, data warehouses, and mission-critical
applications such as enterprise resource planning software (ERPs). Clearly,
managing this growing data is a daunting task for any Information Technology (IT)
department. After all, information only has value if it is stored, can be found,
retrieved, and used. Storage provides the infrastructure to manage data and is
essential for building successful businesses in today’s information age.
Opportunities in Storage
Aside from the sheer growth in data, we expect to see SANs drive storage growth.
Just as Local Area Networking (LAN) drove growth in the 1990s for PCs and
servers and increased the use of computing environments, driving significant
infrastructure investments in the process, we believe Storage Area Networking
(SANs) will increase the use of data and drive storage growth over the next decade.
SANs will increase the Below in Figure 1, we have illustrated PC growth in the 1980s and ‘90s. Growth
use of data and drive significantly accelerated throughout the 1990s, due to (we believe) the advent of
storage growth over the
next decade, in our view.
Local Area Networking (LANs). Users increasingly demanded more from IT
infrastructures in order to remain competitive and/or get a competitive edge. These
IT investments enabled greater productivity and more efficient use of companies’
human capital.

6
The SAN Book III – October 22, 2001

Figure 1. PC Unit Shipments (1981–99)

60,000

Intel
Pentium

50,000

Windows Windows Windows


3.0 3.1 95

40,000
PC Units Shipped (000s)

30,000

20,000

10,000

0
81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99
19

19

19

19

19

19

19

19

19

19

19

19

19

19

19

19

19

19

19
Source: IDC, Bloomberg, and Salomon Smith Barney

An argument could be made that the Internet also had quite a bit to do with PC
growth in the 1990s... and we agree. So, we tried to factor this out by looking at
corporate PC growth. Figure 2 below illustrates the growth of PC shipments
between 1995 and 2000 to large institutions, whose computers are mostly attached to
LANs. These large institutions include large businesses, medium businesses, and
the government. We believe this illustration further validates our belief that SANs
will increase the use of data and points to storage growth through the advent of
SANs.

7
The SAN Book III – October 22, 2001

Figure 2. PC Shipments to Large Customers


Units in Thousands

12,000
17.3% CAGR
10,000

8,000

6,000

4,000

2,000

0
1Q95

3Q95

1Q96

3Q96

1Q97

3Q97

1Q98

3Q98

1Q99

3Q99

1Q00

3Q00

1Q01
Source: IDC

Storage Networking and Software: Show Me the Money!


While we believe that many aspects of the storage industry — such as subsystems
— will continue to offer excellent longer-term investments, we expect to see storage
networking and software be the relative outperformers in the group.

Networking
We believe the storage Storage networking is an emerging trend that is going to unfold in the current decade
networking market is (much like the year of the LAN evolved over many years). We believe it is a very
less than 10%
penetrated, going to
important technology alternative to throwing expensive headcount at the ongoing
85%–90%. challenge of managing more data. We believe the storage networking market is less
than 10% penetrated, going to 85%–90%. In other words, you ain’t seen nothing
yet. Dataquest projects the storage networking market will grow at a 67.0% CAGR,
to $16.9 billion in 2005.
We expect Brocade to continue leading the industry’s growth with its dominant
market position and Inrange to be the market’s up-and-comer by focusing more on a
total solution in the high end with its unique direct sales force. We also believe that
McData will continue to put up solid growth with its end-to-end product focus,
covering both the high-end and low-end markets.
We expect to see excellent growth from the HBA players QLogic, which also has a
switch business that has chalked up some impressive growth, and Emulex, which
has been aggressively penetrating new market segments and customers.

Software
Storage software is an evolving and expanding market that we expect to be an
industry outperformer. We believe the storage software market is undergoing
consolidation (which we would have characterized as “rapid” if IT spending were

8
The SAN Book III – October 22, 2001

healthier). That is to say, the three leading storage software players control a
meager 57.9% of the market with unnamed white box vendors capturing 15%–20%
(EMC with 25.5%, Veritas with 16.3%, and IBM with 16.1% market shares in 2000
according to Dataquest).
We expect EMC and Veritas to continue increasing their respective positions in this
segment; and consequently, to also continue to intensify their head-to-head
competition. Also, we wouldn’t count out Legato, which has a new management
team in place and has been making strides with its partnering approach to the
marketplace.

Subsystems
Storage subsystems are at the heart of storage and we expect that they will continue
to be the major influence in the broader storage industry. We expect EMC to
continue to dominate this segment, although other high-end suppliers, such as IBM
and Hitachi Data Systems (HDS), should continue gaining traction as well. We
expect the mid- and low-end markets to continue to be led by Compaq (which is
scheduled to be acquired by Hewlett-Packard) and NetApp (the industry’s clear
NAS leader) and believe that Dell and Sun Microsystems will play more important
roles going forward.
Commodity?
We find it hard to believe We do not see storage becoming a commodity and find evidence of this at recent
that an industry with this industry events. EMC’s August 1-2, 2001 analyst day (with a full day dedicated to
much innovation and
this much end user
technology tutorials) attracted over 300 investors. Brocade’s first storage
interest in learning about networking conference on June 11–13, 2001 attracted over 500 end users and
new developments is partners. Veritas’s VISION 2000 on October 24, 2000 attracted over 2,500 people,
quickly on its way to (mostly end users and partners: up by about 1,000 from a year earlier and 500 from
becoming a commodity.
two years earlier) to learn about its new direction into storage networking.
In short, we find it hard to believe that an industry with this much innovation and
this much interest in learning about new developments is quickly on its way to
becoming a commodity. Commodity markets are the result of mature industries in
which there is very little product differentiation. We do not see storage moving to a
commodity market for many years.

Market Projections
We believe data storage is one of the most attractive long-term infrastructure plays.
According to Dataquest projections, the overall storage market for multi-user servers
is expected to grow from $36.0 billion in 2000 to $92.0 billion in 2005 in aggregate,
representing a 20.7% CAGR (2000–05).

9
The SAN Book III – October 22, 2001

Figure 3. Storage Industry Projections

CAGR
(Revenue in $millions) 1998 1999 2000 2001 2002 2003 2004 2005 2000-2005
RAID $22,287 $25,593 $29,424 $29,841 $33,176 $38,756 $47,080 $58,399 14.7%
Internal $4,762 $5,818 $7,064 $7,269 $7,223 $6,987 $6,427 $5,704 (4.2%)
JBOD $5,083 $5,704 $4,582 $3,887 $3,162 $2,459 $1,824 $1,289 (22.4%)
External $12,093 $13,491 $16,330 $16,735 $19,270 $23,963 $31,327 $41,063 20.3%
NAS $349 $580 $1,448 $1,950 $3,521 $5,347 $7,502 $10,343 48.2%
Networking $159 $518 $1,300 $2,140 $3,610 $7,360 $13,330 $16,880 67.0%
Software $2,890 $4,088 $5,252 $6,641 $8,456 $10,709 $13,396 $16,729 26.1%
Total Projection $25,336 $30,199 $35,976 $38,622 $45,242 $56,825 $73,806 $92,008 20.7%
YoY Growth -- 19.2% 19.1% 7.4% 17.1% 25.6% 29.9% 24.7% --

Source: Dataquest (August, 2001)

Storage Spending Quickly Outpaces Server Spending


We anticipate storage spending growth, excluding software, to outpace server
We foresee a corporate spending growth by more than 3:1. Dataquest projects storage spending, excluding
spending focus on capital software, to reach $60.4 billion by 2004 (representing an 18.4% CAGR) and server
expenditures, like storage,
that generate high ROIs
spending to approach $87.8 billion by 2004 (representing a 6.6% CAGR, 2000–04).
through cost reductions. We attribute this to a corporate spending focus on areas such as storage that generate
high ROIs through cost reduction and increased operational efficiencies, and
leverage corporations’ most valued asset, information.
Note: Below, in Figure 4, Dataquest is including internal and external-based RAID,
NAS, and networking equipment. Software is not included as it is in Figure 3.
These numbers are taken from a different report so they do not match other tables.
Figure 4. Total Worldwide Server and RAID-based Storage Revenues
($ millions)

100,000
90,000 Storage CAGR: 18.4% Server CAGR: 6.6%
80,000
70,000
60,000
50,000
40,000
30,000
20,000
10,000
0
1999 2000 2001E 2002E 2003E 2004E

Total RAID-based & Networking Storage Revenue Total Server Revenue

Source: Dataquest (July, 2001); and IDC

10
The SAN Book III – October 22, 2001

Data Growth: Doubling Every Year for the Foreseeable Future?


A recent University of California, Berkeley research study pronounced a mega-trend
in information growth, forecasting that the amount of new digital information will
“double every year for the foreseeable future.” Close, but we do not expect this type
of growth in 2001. After talking extensively to storage end users we think data
growth has slowed in 2001 from this trend of 100% growth to about 50%–60%,
reflecting a slowdown in business transactions and new application spending
combined with increased capacity utilization rates. In 2002 we believe information
will almost return to its former growth rate of 100% as IT spending returns, more
new technology applications are implemented, and business transactions resume a
more normal pace. Within our Real World case studies section, we highlight various
organizations and their expected information growth in 2001 and beyond.

Storage: Where Are We and How Did We Get Here?


We expect storage to We expect storage to continue to be an outperforming growth industry. We believe
continue to be an storage is an excellent longer-term infrastructure play. The economic realities in
outperforming growth
industry; however, the
2001, however, have negatively affected the storage market in much the same way
economic realities in as many other technology industries, which should have a dampening impact on
2001 have negatively growth in the near term. Longer term, we expect storage to resume above-average
affected its near-term
growth.
growth.
In Figure 5, we track recent events and how we expect the storage industry to
rebound.
Figure 5. Data Storage and the Economic Pullback

*
tion)
o r p rojec
i
R (pr
CAG nic &
N e w th
24.9% Orga Data Gr
ow
i on
$ Appl
ic a t
20.3%
CAG
R*

th
ata Grow
Flat to declining Organic D rw ay
unde
storage growth Recovery
Resuming
‘normal’
industry
Impact of an growth rate
economic pullback

Time
*Dataquest’s 5 year CAGR

Source: Salomon Smith Barney

11
The SAN Book III – October 22, 2001

The Storage Market: Where Are We?


We expect the storage industry’s growth rate in 2001 to slow to about flat year over
year, down from our prior estimate of about 25% growth.
Based upon feedback from both storage companies and end users, we see both
positive and negative trends for the storage market:

Positives:
➤ In this soft economy, storage companies have not been plagued by excessive
inventory levels similar to that of other technologies (i.e., PC and server
markets).
➤ Data is still growing from mission-critical applications, the Internet, video
streaming, and e-mail. Information growth varies widely across these
applications, but sometimes tops 100% per year in mission-critical applications
like CRM, ERP, and data warehousing.
➤ Pricing of software and storage networking remains somewhat stable despite
increased competition.
➤ We’ve uncovered some large deals and believe storage networking is the next
high growth area within the storage market. We expect IT executives to direct
their limited budgets toward technology investments, such as SANs, that
generate high ROIs within a short payback period.

Negatives:
➤ Storage deals are requiring more top management sign-off, pushing out the
length of the sales cycle.
➤ In the bull market economy, organizations overbought storage capacity in
anticipation of high growth. Today, because of the economic downturn, we are
noticing companies defer storage purchases, by first using up their excess
storage capacity.
➤ We are seeing IT managers postpone storage purchases by increasing their
storage utilization rates. By using say, 85% of available storage rather than
50%, an organization can postpone its storage expenditures.
➤ Organizations are implementing tighter controls around their storage purchases
and are enforcing cost charge-backs to the business units. According to one
source, a year ago, storage purchases seemed to be unrestricted, but today,
management is asking users to order only what is necessary.
➤ Europe does not appear to be a viable source of growth for storage companies in
the short term due to economic conditions.
We believe there are three essential reasons storage is an important future spending
area for IT managers:
1 Spending is increasingly being focused on capital investments with the highest
ROI (return on investment). For example, investments in networked storage
infrastructures, such as SANs, are generating cost savings from increased
capacity utilization, reduced hardware requirements, and lower labor costs. Our

12
The SAN Book III – October 22, 2001

real world case studies confirm this. For example, Intuit witnessed ROIs ranging
from 123% to 525% on its SAN solutions, according to a KPMG consulting case
study.
2 High-end applications like ERP and CRM continue to produce valued
information. As the critical information generated by these applications build,
organizations will use data storage management to leverage information into a
competitive advantage.
3 We believe B2B expansion and corporate webification of larger, more
established companies should continue, ensuring demand for high availability
systems. Storage solutions ensure 24x7 availability of critical data, resulting in
greater customer loyalty and retention.

13
The SAN Book III – October 22, 2001

Demand Drivers: The Real World


➤ Get a grip on the real storage market from our detailed conversations
with many different end users ranging from Inuit to the San
Francisco Giants.

Look at Me, I Need Storage!


Let’s look at a real life, Before we dive into our storage demand drivers analysis, let’s start with an example
personal data storage of real data growth that everyone can understand. Recently having gone through a
crisis.
storage crisis myself, I thought it would be insightful to share the experience. As I
go through the example, think about how this applies to an IT manager and how it
might relate to their experiences.
I recently bought a digital camera that holds 64 MB (megabytes) of memory per disk
or roll of film. I have found that with my digital camera I take many more pictures
due to the decreased cost of development. Below in Figure 6, Dataquest’s research
illustrates the attraction of digital photography.
Figure 6. Digital Camera Usage

What Happens To Picture Taking Activity After


Purchasing A Digital Camera?
Other
Response
35%

Take More
Take More Digital Pictures
Traditional & Less
Pictures Traditional
15% 50%

Source: Dataquest

After going through a few film cards and downloading them onto my personal
computer with a 20 GB (gigabyte) hard drive, I realized that at the rate I was going, I
would fill up my disk storage space in a year (I already have a lot of other
applications on my PC). At that time, I would either have to start deleting data or
pictures, or buy a new computer. As an alternative I could have replaced the hard
drive (I can easily buy up to 80 GB today), but that is quite painful and there is no
telling what would happen if I let loose with a screwdriver on my computer to pop
the lid off and start ripping out the drive and installing a new one. What would
happen to my old data? How do I transfer data from the old disk drive (out of the
system) to the new disk drive? Wouldn’t the computer have to be reconfigured? For

14
The SAN Book III – October 22, 2001

Pete’s sake, I have a tough enough time upgrading my software...hardware is out of


the question.
Illustrating the trend I soon realized that attaching external storage to my computer was the only viable
towards 1)external solution to create independent scalability (note the trend toward external storage). I
storage,
thought about tape or CD read/writable, but the thought of cluttering my home with
2) online data or data another piece of furniture that held CDs or tapes (i.e., removable media) was not too
availability,
appealing. In addition, I wanted to have my pictures organized and easily accessible
(note the trend toward online data or data availability). So, I looked into a NAS
disk drive appliance.
3) increased spending on From Quantum I could get a SNAP Server 2000 for $1,400 or from Maxtor I could
storage relative to get a MaxAttach 3000 for $950, both with 80 GB of storage. These storage
traditional computing,
and
appliances would cover my storage expansion for quite some time, even at the torrid
pace I was using my new camera (note that my computer would cost less than $900
today and I was spending $1,000 on external storage; i.e., the trend toward
increased spending on storage relative to traditional computing). But, I was leaving
myself in a vulnerable position. What would I do if I was robbed and my computer
was stolen? Sure the insurance would replace my computer, but what about my data
(pictures, etc.)? What would happen if there was a fire? Or an earthquake? Or
locusts?
4) multiple copies; i.e., I needed a backup copy, but without adding another piece of furniture, oh, and
data replication.
without taking too much time (note the shrinking backup window and need for two
copies of data). So, I looked into Snapfish.com, an online provider of film
developing, printing, and storing photos. They could handle a second copy of my
photos (note the trend toward multiple copies; i.e., data replication) and provide me
with other services too. That way if I were robbed I could go to Snapfish.com to get
my pictures online and I would still have my own copy on my computer. Perfect!

15
The SAN Book III – October 22, 2001

Figure 7. Online Photo Sites

Source: PC Data

Conclusions
➤ Data is important and increasingly so; for example, if my pictures are this
important to me, imagine how important a company’s most important asset
(data) is.
➤ Data is growing in many new areas and the need to store it, replicate it, and back
it up is increasing exponentially.
➤ Spending is shifting toward storage.
➤ Storage is moving to be external and networked (i.e., accessible).
➤ Information technology managers want their data online and replicated (i.e.,
more than one copy).
Now Check This Out!
Snapfish.com saw its storage grow more than 28 terabytes in the first year of
business and expects it to double again in 2002. Snapfish has several competitors,
including AOL and Kodak. If each competitor witnessed data growth comparable to
what Snapfish expects, the combined storage demand would increase by at least 500
terabytes annually. That alone could fit on EMC’s top 25 customer list.
Driving the point home: if AOL’s “You’ve Got Pictures,” an online digital photo
service, gets just 5% of its customers from existing subscribers and each customer
takes roughly 25 pictures per month and stores it using AOL, we estimate that could
drive over 200 terabytes of storage demand, annually.

16
The SAN Book III – October 22, 2001

And Then Add This!


IDC projects worldwide Of course AOL is increasing its subscriber base, in fact it doubled over the last two
digital camera shipments and a half years, to (now) 30,000,000 users, according to AOL. Also digital camera
could reach 38.7 million
by 2005, representing a
sales have just begun to ramp. In fact, IDC projects worldwide digital camera
20.7% CAGR (2000–05) shipments could reach 38.7 million by 2005, representing a 20.7% CAGR (2000–
05).
Figure 8. Worldwide Digital Cameras

45,000 $14,000
40,000 $12,000
35,000

Value of Shipments ($M)


Value of Shipments (Right Scale)
$10,000
Shipments (000s)

30,000 10.4% CAGR 2000-2005


25,000 $8,000
20,000 $6,000
Unit Shipments (Left Scale)
15,000
20.7% CAGR 2000-2005 $4,000
10,000
5,000 $2,000

0 $-
1998 1999 2000 2001E 2002E 2003E 2004E 2005E

Source: IDC (May, 2001)

And that is just digital photo imaging; medical imaging is also an expanding storage
market. The University of California, Berkeley estimates that there are roughly 180
million chest x-rays and mammograms taken annually in the United States. At
roughly 8 MB per x-ray, that represents potential annual storage demand of over
1,400 TB. Again this annual storage demand is equivalent in size to three of EMC’s
top customers.
While we focused our discussion above on potential digital photography storage
demand, we could have chosen many other areas, including:
➤ mission-critical applications such as ERP (enterprise resource planning) and
CRM (customer relationship management)
➤ e-mail and the growing popularity of instant messaging
➤ wireless content and handheld devices, including Research In Motion pagers
➤ digital audio (digital music downloading)
➤ digital data archiving including e-mail archiving
➤ streaming video
➤ set-top boxes, Web TV and other digital TV services
We believe the conclusions we summarized above apply to all of these, and other,
emerging technologies.

17
The SAN Book III – October 22, 2001

Storage: The Real World


To hear about real IT We thought it would be helpful to provide some insight into our regular
managers’ real world conversations with IT (information technology) executives; to hear their real world
experiences, read on.
stories on what is driving information growth and the demand for storage.
Throughout this section we provide excerpts from our conversations with IT
executives from a variety of industries.
Note: While many IT managers cited data growth rates to be below 100% going
forward, most of these are only organic projections and do not include new
application growth, which we believe counts for the bulk of storage growth in a
healthy market.
Figure 9. Our Featured Storage End Users

Other
Storage
End-Users

National Climatic
Data Center (NCDC)

Source: The Internet and companies

There is a consistent theme to what we are hearing from a diverse group of storage
end users including those shown in Figure 9 and others remaining anonymous
(which we have labeled Company X and Company Y):
1 Data is growing. The biggest drivers of data growth are mission-critical
applications, the Internet, video streaming, and e-mail with large media
attachments. Growth rates vary from 30% to 100%-plus in some of the more
critical applications.
2 Data growth, however, is slower in 2001 than 2000. A slowing economy
means fewer transactions and declining new application deployments.
Consequently, storage capacity bought in 2000 lasted longer than expected,
reducing 2001 storage expenditures. We believe information growth in 2001
could fall in the range of 50%–70% (for capacity, not revenues). Beyond 2001,
we expect data to return to its former trend of 100% growth per year.
3 Storage capacity utilization is rising. IT managers are using up their excess
storage capacity, by squeezing as much out of their IT infrastructure as possible.
This has resulted in delayed storage purchases in 2001. Storage Area Networks
(SANs) also lead to higher utilization rates because of their centralized

18
The SAN Book III – October 22, 2001

architecture. Most end users expect to maintain these higher levels of storage
utilization beyond 2001, which has a dampening (but not crushing) effect on
storage growth going forward.
4 Price per megabyte of storage is declining. End users are telling us they are
seeing (or forecasting in the case of post 2001) price declines of approximately:
Pre-2001 35%
2001 50%
After 2001 40%
Source: Salomon Smith Barney

5 SANs (storage area networks) generate high ROIs — largely from labor
savings, floor space reduction, and reduction of file server hardware and
maintenance to support that hardware. The decision to migrate to a SAN from a
DAS (direct-attached storage), however, must be driven by a strong business
purpose.
6 SANs target market is large. Several end users we talked with (many not
mentioned by name) were small businesses with less than 8 TBs of data. What
interested us, was that almost all of them were using or were planning to
implement SANs. This could mean broader SAN adoptions than most anticipate.
7 Data replication will increase in 2002 and beyond. Driven by declining
storage and bandwidth prices, and advances in software technology, we expect
demand for replication to continue to grow well into the future. (Veritas’s
Replicator and EMC’s SRDF are the two leading products in this area).
8 Companies’ information infrastructures are closely protected assets.
Throughout our discussions, we noted that an increasing number of IT managers
either could not discuss their storage environment or did not want to be
mentioned by name for competitive reasons. We find the fact storage
infrastructures are being treated as a tightly protected corporate asset, as a very
encouraging sign for future emphasis on spending in this area. Increasingly,
things such as discussing how many terabytes a company has, how they are
managing it, how they are growing it, and what they expect to do going forward
to better manage and use their information, is becoming a secret (i.e., a
competitive advantage).
Below we provide more details of our conversations with IT executives.

Intuit: A SAN Success Story


Within two years of Intuit aids in the filing of over 32 million U.S. taxpayer returns and has 60 terabytes
implementing a SAN, of data, growing at roughly 50%–60% per year; with some of its mission-critical
Intuit realized a return on
its investment that
applications growing at more than100% per year. Intuit is a great SAN success
ranged between 123% story: The company implemented a SAN two years ago and realized a return on its
and 525%. investment that ranged between 123% and 525%, depending on the storage solution.
Payback period ranged from two to nine months! (see Figure 10 below).
Now sit back and imagine you’re an IT executive of a company that is responsible
for the online submission of tax data for millions of U.S. taxpayers. Your busiest

19
The SAN Book III – October 22, 2001

time of year is in the month leading up to the tax filing deadline and you have a
narrow window of opportunity to perform your system backups. Unfortunately, the
actual time required to do the backup keeps increasing due to system load. Just two
days before the critical April 15 tax deadline it takes 26 hours to complete the
backup. This happened to Intuit just three years ago and during those 26 hours, the
system was unavailable to its online customers.
Because of this event and the advances in technology, Intuit implemented a robust
centralized storage area network (SAN), one that would provide it with the system
reliability, availability, and scalability the company was not getting from its direct-
attached storage (DAS) infrastructure. The old DAS environment had subsystems
linked to over 1,000 servers and was difficult and inefficient to manage. KPMG
Consulting, which conducted an extensive ROI (return on investment) analysis on
Intuit, found that Intuit’s SAN ROI ranged from 123% and 525%, as shown in
Figure 10 below.
Figure 10. Intuit’s SAN ROI
Return on Payback
Solution Investment Period
Storage Consolidation 296% 4 months
Backup/Restore 123% 9 months
System Availability 525% 2 months
Source: KPMG Consulting and Brocade Communications Systems Inc.

Here we highlight the key improvements in Intuit’s IT infrastructure as a result of


the SAN implementation and the factors fueling the high ROIs:
➤ Capacity utilization: Intuit’s storage capacity utilization improved to 85%
throughout most of the SAN infrastructure. This was a significant increase over
the 50% utilization rates found in the pre-SAN environment, according to
Intuit’s IT executives.
➤ Reduced server hardware requirements: Intuit reduced its server hardware
and maintenance costs by 50% in its SAN environment, according to KMPG
Consulting and Intuit executives.
➤ Increased system reliability and availability: In the SAN environment, Intuit
now conducts a daily backup of its system without disrupting the Internet
customers. In the past, backups could only be performed once a week, because
anything more than that created too much system downtime.
➤ Labor cost savings: Intuit reduced the number of employees managing the
backup and restore environment from five to six people (working all night) to
two people (leaving by 8.00 p.m.). In the “old days,” employees joked that they
needed roller skates to keep up with the task of changing the storage tapes. The
staff now manages 5x–10x more data in the post-SAN environment with much
less effort.
➤ Faster Testing and Development: In the SAN environment, multiple copies of
data can be made available for production and testing by software developers.

20
The SAN Book III – October 22, 2001

This has increased the time to market for new software applications, a key
advantage in a competitive business environment.

“Company X” — A BIG Corporation with Terabytes


Around the World
Company X has Company X’s product development business unit, which tests new software
consolidated its data enhancements for its mission-critical billing system, has consolidated its data centers
centers by 50% and cut
the workload of its IT
by 50%, and cut the workload of its IT staff by one-third because of its migration to
staff by one-third a SAN, according to conversations with Company X IT executives and a KPMG
because of its migration Consulting study.
to a SAN.
As an early adopter of a SAN in its product development group, Company X
provides great insight into the benefits of a SAN. Below we highlight key points
from our conversations with Company X’s IT executives regarding its SAN
environment:
➤ Consolidation of testing and production environments: The flexibility of the
SAN architecture allowed Company X to consolidate its testing and
development environments from four to two locations. The group witnessed a
significant reduction in the hardware and management costs previously needed
to replicate data from these various sites.
➤ Multiple storage vendors: SAN infrastructures opened the door for Company
X to use multiple vendors resulting in greater pricing leverage.
➤ Reduction in server hardware: SANs enabled Company X to consolidate its
servers and reduce future server expenditures by an estimated 50%. As a result,
the IT staff workload was cut by almost one-third, allowing staff to be
redeployed to more strategic areas.
➤ Data center — physical space cost savings: As a result of the SAN
architecture, Company X reduced its file servers by about 50%, and
considerably reduced the amount of floor space needed in the data center.
➤ Capacity utilization: In the SAN environment the business unit of Company X
will typically strive for a 75% utilization rate (25% of storage capacity on the
shelf), a significant improvement from DAS levels, which were closer to 50%.

“Company Y” — A Company Within the Technology


Hardware and Equipment Sector
We talked to one large company within the technology hardware and equipment
sector with more than 500 terabytes of managed data, 30% of which is housed in a
SAN. To give some perspective of the magnitude of data, five hundred terabytes is
equivalent to all of the printed material in 50 Libraries of Congress or 25,000 miles
of bookshelves. Prior to 2001, the company’s data storage roughly doubled every
year, but this year growth is expected to be flat, because of the slowing economy
(fewer transactions) and cost-cutting measures. The company expects storage
expenditures to be cut 60% in 2001 and to be more disciplined about storage and
hardware spending in 2002 and beyond.

21
The SAN Book III – October 22, 2001

Here are some details of our conversations with this company’s senior IT
executives:
➤ Backup reliability: Imagine the emotional strain on an IT department if data
backup and replication processes fail every night. This is what the company
faced in its nightly data migration to the supply chain management reporting
group. After the problems were escalated, the company looked to EMC’s
TimeFinder to perform backups quickly, frequently, and without disruption.
Today, 22 months after implementing TimeFinder, systems run virtually
uninterrupted.
➤ Data growth is slowing in 2001: Prior to 2001, the company’s storage roughly
doubled every year, but this year it expects growth to be flat, because of the
slowing economy (fewer transactions) and cost-cutting measures (no new
application deployments and a reduction of some older data). The company
expects to be more disciplined about storage and hardware spending in 2002 and
beyond.
➤ Storage spending in 2001: The organization expects to cut storage spending by
60% in 2001, as part of its cost-cutting program. To do this the company will
increase its storage capacity utilization, clean up storage files by deleting
outdated and duplicate files, and restrict storage capacity for each business unit.
➤ On the technology wish list: reliable and proven virtualization software from a
reputable storage vendor.

NASA: Rocket Scientists Need Storage Too


The Goddard Space Flight Center (GSFC) is a NASA field center dedicated to
operating unmanned spacecraft and managing many of NASA’s earth observation
missions. In 1998, GSFC launched the Earth Observing System (EOS) which is a
series of spacecraft and supporting data systems that study the interactions of the
atmosphere, land, and oceans. EOS will develop a 15-year database to focus on the
Earth’s climate change. After reading about this program and hearing Veritas
comment about the strong growth in sales to the government sector, we thought it
would be interesting to talk to NASA about its storage infrastructure at GSFC.
Below are some excerpts from our conversations with a senior computer scientist
and head of the SAN study at GSFC.
➤ Data growth: GSFC currently has about one petabyte (PB) of data, and in
2001, data is expected to grow at a rate of 250 terabytes (TBs) per year or
roughly a 25% annual growth rate. In 2002, IT executives expect GSFC will
generate more than 500 TB of data per year, increasing to a 40% data growth
rate.
Most of the current growth can be attributed to the EOS system, which generates
about 1.2 TB of data per day over seven distributed active archive centers (of
which GSFC is one) throughout the United States. Over a 15-year period, the
EOS system will generate significant amounts of data, which the U.S.
government intends to keep forever for future scientific observation. Figure 11
below details the storage demand that could be generated by the EOS system

22
The SAN Book III – October 22, 2001

through 2010; assuming an annualized 50% growth rate. Note: This assumes no
new application deployments.
Figure 11. Annual Storage Demand for GSFC EOS Satellite System

Projected EOS Annual Storage Demand


(in terabytes)

38,443
40,000
50% CAGR (2002-2010)
30,000

20,000 17,086

10,000 5,063
1,500
0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: NASA and Salomon Smith Barney

➤ Disk versus tape storage: Today, roughly 15% or almost 150 terabytes of
GSFC’s data is stored on disk with the remaining 85% housed in a tape archive.
The disk storage is the primary archive and GSFC uses caching to improve
access time to critical data. As disk prices decrease, GSFC expects to migrate
more data to disk from tape, which will improve access time and allow
researchers to perform data mining/warehousing on archived data.
➤ Disk versus tape: GSFC is weighing the cost versus performance benefit of
tape, enterprise performance disk, and IDE disk (desktop). GSFC will likely use
all three solutions to manage its data economically and efficiently. For GSFC,
tape is economical, reliable, portable, and secure — great for storing data which
is accessed less frequently. Enterprise performance disk is the best-performing
but most expensive of the options. It can be used very effectively in connection
with tape, especially when disk is used only for storing critical data requiring
fast access times. Finally, GSFC is also considering using IDE disk or desktop
disk, an option that has higher performance capability than tape and is about
one-tenth of the price of enterprise disk. Of course, using three different
solutions increases the complexity of the storage infrastructure. We will
continue to monitor GSFC to see how its storage infrastructure unfolds.
➤ SAN implementation: GSFC has recently begun a SAN pilot implementation,
and is looking to the SAN to provide high speed file sharing, high availability,
and reliability. Initially GSFC will house only a small fraction of its disk data
on the SAN, but if the first phase goes as planned, GSFC expects that a large
portion of its disk-based data would eventually be migrated to the SAN. In the
SAN environment, one of GSFC’s biggest concerns is data security.

23
The SAN Book III – October 22, 2001

Lawrence Livermore National Laboratories: from the


Human Genome Project to Nuclear Warheads
Lawrence Livermore National Laboratories (LLNL) is a U.S. Department of Energy
defense programs laboratory that uses advanced technology on projects to protect
national security: from human safety (human genome project) to maintaining the
enduring nuclear stockpile.
Here are some highlights of our conversations with LLNL’s IT managers:
➤ Storage capacity growth 1997–2001: Currently LLNL has over 250 terabytes
of disk storage and a tape archive with over 2,000 terabytes of capacity,
representing an estimated 202% and 151% CAGR (1997–2001) for magnetic
disk and tape storage capacity respectively. Note the trend of higher disk growth
rates compared to tape.
Figure 12. Lawrence Livermore National Laboratories’ Projected Disk and Tape Storage Demand

12,000 CAGR 2001-2003: CAGR 2003-2005:


10,000
10,000 Disk: 85% Disk: 139%
Tape: 71% Tape: 58%
8,000
Terabytes

6,000
4,000
4,000
2,000 2,000
2,000
1 20 250 350
0
1997 2001 2003 2005

Disk Storage Tape Storage

Source: Lawrence Livermore National Laboratories and Salomon Smith Barney

Through 2005, LLNL ➤ Storage capacity growth expected for 2001–05: Through 2005, LLNL
expects to increase its
expects to increase its disk and tape storage to 2,000 and 10,000 terabytes,
disk and tape storage to
2,000 and 10,000 representing a 68% and 50% CAGR ( 2001–05), respectively.
terabytes, representing a
68% and 50% CAGR
➤ Storage capacity utilization: In the disk environment, storage capacity
(2001-05), respectively. utilization varies between 50% and 85%, depending on the resource.
➤ Drivers of information growth: Programs simulating various physics
scenarios are generating tremendous amounts of data, roughly 95%–98% of
LLNL’s total storage usage. Maintaining various inventories is driving 159%
and 117% CAGRs (1997–2005) in LLNL’s disk and tape storage, respectively.
➤ Most significant problems in moving/sharing data: bandwidth between
multiple large computing resources, reliability of file systems, followed by
manageability.
➤ Benefits of advanced storage architecture: With the advent of NAS
(networked attached storage) and NFS (network file systems) servers, LLNL has

24
The SAN Book III – October 22, 2001

been able to provide common home directories for its scientific computing users
on all platforms. In the future, LLNL expects SAN/NAS capabilities will allow
them to have an enterprise-wide file system that scales in terms of bandwidth,
capacity, and manageability (resolving its most significant problems, see above).
LLNL plans to deploy this type of technology in a three- to four-year timeframe.
This architecture will allow LLNL to have only one copy of the simulation data
and present that data on the computational engines, visualization resources, and
long-term tape archival resources.
➤ Data storage budgets are increasing: Currently, storage spending, including
disk and tape, is about 25% of LLNL’s total IT budget. LLNL expects the
storage budget to increase in 2002 to 35%, to support a new super computing
device.

iVillage: Effective Centralized Tape Backup


iVillage is a leading online women’s Web portal which has storage challenges that
are representative of many small to mid-size companies. The company manages a
little over 1.5 terabytes of data.
iVillage went from ➤ Centralized tape backup helps: After implementing an automated tape robot
spending one to two in its SAN environment, iVillage was able to consolidate its tape libraries and
hours per day for backup
to one to two hours per
cut back on its storage maintenance from one to two hours per day (server by
month simply by server, one by one) to roughly one to two hours per month (in aggregate).
aggregating storage
resources. ➤ Data availability/reliability: improved significantly because of the tape
consolidation. Today, backup failures occur infrequently compared to nightly
failures in the old decentralized environment.

Duke University: Storage Is the “Killer App”


For Duke University, storage is a driving factor in the research projects it undertakes
and not the other way ‘round. New and advanced research projects would not be
possible without advanced storage architecture (i.e., SAN) and the declining price of
storage. Duke University has about 15 terabytes of data spread between its
academic and medical center units.
We talked to Duke’s Associate Chief Information Officer about recent SAN
implementation and the drivers of information growth at Duke University:
➤ A SAN, backed up by a huge tape library: Duke recently implemented a
SAN which has about 3 terabytes of disk capacity backed up by an IBM tape
storage device that has 42 terabytes of capacity and an automated tape robot.
Benefits of this centralized storage setup include greater reliability of the nightly
backup process and a reduction in the physical size of the data center. The
centralized SAN storage environment also allows researchers to share pooled
information quickly and reliably.
➤ Data growth drivers: The biggest drivers of information growth at Duke
University are genomics data, digital imaging from research projects, and e-mail
with rich media attachments.

25
The SAN Book III – October 22, 2001

➤ Storage — driving research initiatives: A recent brain imaging/cognitive


neuroscience project requires researchers to share access to thousands of digital
brain images from test subjects. The sheer mass of required data storage,
potentially up to 45 terabytes, would have deterred researchers from undertaking
the project if it were not for less expensive storage costs and Duke’s advanced
SAN infrastructure.

Snapfish.com: Digital Imaging Is Doubling Every Year


Snapfish expects data to Snapfish.com, the online provider of film developing, printing, and storing photos,
double in 2001 and mentioned in our earlier digital photography anecdote, saw its storage grow 28
beyond.
terabytes in the first year of business, and expects data to double in 2001 and
beyond. Here are some of the storage insights we got from Snapfish’s Chief
Technology Officer:
➤ Digital Imaging Storage Market: Snapfish has several competitors including
AOL and Kodak. If each competitor witnessed comparable growth to what
Snapfish expects in 2002, storage demand would increase by more than 500
terabytes annually.
➤ Storage utilization: Snapfish typically maintains high levels of storage
utilization, up to 90% in the storage area network.
➤ Declining storage prices: Despite Snapfish’s expectation that data could
double next year, IT executives expect storage spending to be flat year over year
because of rapidly falling storage prices. Currently, storage represents roughly
70%–80% of Snapfish’s IT spending.

National Climatic Data Center


The National Climatic Data Center (NCDC) has over 550 terabytes of data in its
international archive facility for meteorological data, ranging from data collected by
Thomas Jefferson and Benjamin Franklin to data collected utilizing the most modern
collection systems and satellites. By the end of 2002, the NCDC expects to have
one petabyte of data in its facility.
Here is what we learned in our conversations with NCDC IT management:
➤ Data Growth: At NCDC, data is growing at a rate of about 40%–50% per year
and is expected to continue at this rate through 2005. Almost 100% of the
facility data is backed up to tape storage.
➤ A SAN Infrastructure: Despite a limited budget, NCDC will implement a
SAN in 2001 in order to provide its scientists with fast and reliable access to the
most frequently used, mission-critical information (roughly ten terabytes of
data).
➤ Robotic Tape Libraries and Data Replication: Roughly 90% of NCDC’s
data is stored on tape. NCDC also plans to implement an automated tape library
along with the SAN, to ensure reliable data backup and lower storage
maintenance costs. The NCDC currently replicates 100% of its new data and by

26
The SAN Book III – October 22, 2001

2005 expects to have replicated all of its data (projected at one petabyte) to tape
for disaster recovery purposes.
➤ Hierarchical Storage Management (HSM) Software: NCDC uses a
hierarchical storage management software to determine what data should be
deemed critical and stored to the magnetic disk drives that provide faster data
access. As the price per megabyte of magnetic disk declines, the NCDC expects
to move a greater proportion of data to disk.

The San Francisco Giants: 40,000 Fans and Seven


Terabytes
The San Francisco
Passing through the electronic, ticket-scanning turnstiles at the San Francisco
Giants have Giants’ Pacific Bell Park, we knew that our nation’s oldest pastime was now
approximately 7 generating and storing significant amounts of critical data. But what we found in
terabytes of data
our conversations with the Giants’ senior IT executives impressed us.
capacity to support
video streaming and an The Giants, one of Major League Baseball’s most tech-minded franchises, have
online ticket trading
more than two terabytes of data growing at a rate of approximately 0.5 to 1 terabyte
platform.
per year with seven terabytes of total storage capacity. Two seasons of digital video
game footage used for coaching and scouting has already generated a little over a
terabyte of data. This number is expected to grow at about 50% per year; in fact, the
Giants expect to store an “eternity’s worth of game footage” in their storage
infrastructure in order to garner a competitive advantage, according to a senior IT
executive.
Figure 13. Barry Bonds of the San Francisco Giants: An Unlikely Storage End User

Source: sfgiants@mlb.com

The Storage Lineup


Enabled by a strong storage lineup, including a RAID 5 (redundant array of
independent disks) configured 750 DVD jukebox (5.2 gigabytes per DVD) and OTG

27
The SAN Book III – October 22, 2001

software, the Giants believe the information collected from their electronic
turnstiles, e-ticket system, and video streaming will give them a competitive
advantage, improve their win/loss record, and bolster their bottom line. With only
seven out of the 30 Major League Baseball parks coming close to the Giants’ level
of technological prowess, we think they are right. As the business value of
information becomes apparent, the other parks are likely to follow suit. At seven
terabytes of capacity each, Major League Baseball parks combined could have
storage requirements similar in size of say, Citigroup, one of EMC’s largest
customers.
Figure 14. Potential Storage Demand for Major League Baseball

250
210 220
200
Terabytes

150

100

50
7
0
Giants MLB (potential) Citigroup
Storage Demand
Source: Salomon Smith Barney

We think the Giants are a great example of how traditional old-economy businesses
are beginning to store large amounts of critical information to gain a strategic
advantage. Who’s next? The NBA? The NFL? With this in mind, we highlight a
few of the Giants’ innovative technologies that are driving their information growth.

1. Video Streaming Is Being Used for Coaching and Scouting


Today from a front-end platform in their clubhouse, players can query digital video
game footage from their last two seasons to analyze their performance. If Barry
Bonds wants to analyze video footage from a past at-bat from multiple camera
angles, he can — within seconds. The Giants’ storage infrastructure ensures that the
players can access archived Giants game footage with high availability and speed.
According to the Giants’ Senior IT executive, “the players won’t wait patiently at
the terminal to obtain the data. They expect the data and technology to be available
within seconds of sitting down, otherwise it has no use to them.”

2. E-turnstiles and E-tickets Are Generating Valuable Data


Today, approximately seven out of 30 ballparks use electronic turnstiles that provide
valuable information about ticket holders. In the server room while the game is
going on, the Giants can see how many fans are in the park, what seats are selling
out fastest, what kind of ticket holders are entering each gate, and the efficiency of
their greeters at each turnstile.

28
The SAN Book III – October 22, 2001

Figure 15. Pacific Bell Park: E-turnstiles, 40,000 seats, and 7 Terabytes Storage Capacity

Source: sfgiants@mlb.com

In addition to the e-turnstiles, the Giants have also implemented an e-ticket system
in their luxury suite area which issues tickets after fans swipe their driver’s license
or credit cards into the system. As a result, the Giants have access to a tremendous
amount of essential demographic information on their luxury box ticket holders. For
example, the Giants can analyze the number of Dodger games a fan attended, the
preferred day of the week, and whether s/he appears to like day or evening games.
According to the Giants’ senior IT executive, “I can definitely envision a world in
the not-so-distant future where we are 90-plus% paperless.”
While there are no definite plans to expand this system on a stadium-wide basis, all
suites and season ticket holders could be using this paperless ticket system in the
next two years. At that point, the Giants could be managing demographic data on
their fans at the level of a credit card company. Enabled by their strong storage
lineup, this information would give the Giants tremendous insight into their
customer needs and would greatly expand their sales and marketing capabilities.

3. Online Ticket Trading


Double Play Ticket Window (DPTW) facilitates online ticket trades for the Giants’
30,000 season ticket holders. This technology is only possible because of the park’s
e-turnstiles, which ensure that no two people are allowed into the park for the same
seat. Through DPTW, the Giants collect valuable information on:
➤ average mark-up on traded ticket sales. For example, a St. Louis Cardinal’s
ticket (i.e., Mark McGwire) might sell at a 70% premium, while the Florida
Marlins ticket trades at face value. Analysis of the data gives the Giants the
flexibility to develop a dynamic or variable ticket-pricing model in the future.
➤ historical ticket demand and pricing — which can be used for scheduling games
to get the highest possible attendance.

29
The SAN Book III – October 22, 2001

Demand Driver Analysis


To recap: First, we outlined our view of the current storage market, the market
projections, and some of the positive and negative trends of the sector. We followed
this with a real world perspective of what is driving information growth and demand
for storage within various organizations.
Now we analyze the most significant factors driving demand in the data storage
market. Since we believe this market is best understood by looking inside a data
center, we relate much our analysis to excerpts from our Citigroup, New York case
study.
Case Study: Citigroup, New York
Citigroup, one of EMC’s top 15 customers, is a good representation of where the
broader storage market is headed. Citigroup’s storage capacity has grown to over
220 terabytes in its New York area technology center.

Computer Storage Demand Drivers


We believe the demand for computer storage in 2001 and beyond is driven by the
following factors:
➤ storage capacity utilization
➤ data replication
➤ high returns on investment in storage technology
➤ information growth from:
— mission-critical enterprise applications
— the Internet (e-mail and e-commerce)
— emerging “killer applications” (for example, video streaming)

Storage Utilization and Data Replication


We believe there is a Below we discuss two conflicting drivers of storage demand:
current trend toward
higher storage utilization Point: Higher Storage Utilization Rates Decrease Storage Demand
rates, which could have
By increasing storage utilization (or the amount of storage capacity used),
a negative impact on the
demand for storage. organizations can postpone their storage purchases. Based upon our conversations
with end users, we believe there is a current trend toward higher storage utilization
rates which will likely have a negative impact on storage demand.
A year ago, it was common for a rapidly growing organization to leave 50%–60% of
its storage capacity empty in preparation for anticipated growth. Internet data
centers, for instance, typically need to rapidly scale their capacity to capitalize on the
sudden success of their site. Today, because of tighter budgets and slowing growth,
IT managers are raising their storage utilization rates to cut back on their storage
spending.
We believe higher storage utilization rates will have a negative effect on the demand
in the near term for storage capacity, for the following reasons:

30
The SAN Book III – October 22, 2001

➤ IT spending has slowed, so organizations are cutting down on storage purchases


by squeezing as much out of their technology infrastructure as they can. For
storage subsystems, this means increasing the utilization rates from an average
of say, 50%, to perhaps 85%, as depicted in Figures 16 and 17.
➤ Storage Resource Management (SRM) software enables organizations to gain
more efficiency out of their storage infrastructure, yielding higher capacity
utilization rates, and fewer storage purchases. We are seeing increased market
penetration for SRM software indicating that storage utilization will be trending
upward. We discuss SRM in more detail in the “Software Applications” section.
Below in Figure 16 we illustrate how an increase in storage capacity utilization to
85% from 50% negatively impacts storage demand. The bottom line for the
hypothetical organization below: storage demand declines by roughly 40% in
response to an increase in storage capacity utilization to 85% from 50% over an
undetermined time. We expect this increase in capacity utilization to be worked
through by the end of 2001, at which time data demand should begin to resume its
normal growth trajectory.
Figure 16. Example of the Demand Effect of Increased Capacity Utilization

4.50
Inflection Point: Storage demand drops as
3.75
Storage Demand (Terabytes)

utilization increases from 50% to 85%. Demand


3.00 Effect:
40%
2.25 d li

1.50

0.75

0.00
Time

50% Utilization 50% to 85% utilization 85% Utilization

Source: Salomon Smith Barney

SANs Increase Capacity Utilization Even Further


Furthermore, we expect greater increases in storage utilization as flexible storage
architectures, such as SANs, show increasing market penetration. We draw your
attention to Figure 17 to illustrate this point. In a traditional direct-attached storage
(DAS) architecture, technical limitations require companies to directly pair storage
subsystems with servers (i.e., server 1 with disk subsystem 1). Unused capacity
cannot be easily moved to servers that require additional storage.
For example, on the left side of Figure 17, disk subsystem 2 is full and cannot
borrow capacity from the other underutilized subsystems. Despite a 50% capacity

31
The SAN Book III – October 22, 2001

utilization in the DAS environment, the organization must purchase more storage if
server two needs additional capacity. The SAN environment depicted on the right-
hand side of Figure 17 has 4 TB (terabytes) of data, but only requires 4.75 TB of
storage. This represents a 40% decrease from the 8 TB of storage required in the
DAS environment, which could not exceed more than 50% capacity utilization.
Figure 17. Demand Effect of Increased Storage Capacity Utilization

DAS: 50% Utilization SAN: 85% Utilization


8 TB
available storage 4.75 TB
available storage

1 TB: 50%
utilization SAN
Clients Clients Manager
Server 1 Server 1

Disk
Subsystem 1 2TB: 100% Disk

LAN utilization LAN SAN Subsystem 1

Server 2 Server 2
Disk
Subsystem 2 1.5 TB: 75% Disk
utilization Subsystem 2

Server 3
4 TB data =
Server 3 Disk .5 TB: 25% 85%
Subsystem 3 utilization utilization

Server 4
Server 4 Disk
Subsystem 4

1) 50% utilization = 4TB data / 8TB of available 1) 85% utilization = 4TB data / 4.75 TB of available
storage capacity) storage capacity
2) Needs of disk subsystem 2 requires additional storage 2) No wasted capacity with room to grow
purchase
3) Excess capacity in arrays 1,3, & 4 cannot cover needs

Result: Storage demand declines 40% with Utilization Increase


Source: Salomon Smith Barney

Real World: In its SAN environment, Citigroup orders additional storage when
capacity reaches 90% compared to 50%–60% in its DAS environment. By simply
expanding utilization rates, Citigroup tapped roughly 35% more storage capacity
without spending a dime.
Counter-Point: Data Replication — Driving Storage Demand
Data replication refers to the duplication or “mirroring” of an organization’s
mission-critical data for disaster recovery purposes, data migration, or business
operations (i.e., development and testing). As data replication increases, so should
the demand for storage. We believe the increasing market penetration of data
replication initiatives should, in the long run, have a profound affect on the storage

32
The SAN Book III – October 22, 2001

market and should offset some of the negative demand effect of higher utilization
rates.
For some organizations, like Citigroup, regulatory rules require the maintenance of
much of the company’s mission-critical information in a disaster recovery facility to
ensure high availability. For others, like Intuit (discussed above), replication of data
ensures faster and more accurate testing of new software enhancements, since
developers can quickly access a reliable and updated copy of data.
We believe that the projected declines in storage subsystem and bandwidth prices
and the advances in data replication software are all pointing to future demand for
data replication.
Note: We explain replication in more detail in the “Software Applications” section.

Data Replication Software Revenues Are Projected to Rise


Data replication software, which is designed to create image copies of data via
mirrors, clones, and snapshots, is projected to grow to $3.3 billion in 2005 from
$967 million in 2000, representing a 28% CAGR (2000–05). This rising demand for
data replication software points to an increase in the demand for new storage
capacity to house new copies of data.
Figure 18. Data Replication Software Revenues

3,500
3,000 28% CAGR (2000-2005)
2,500
$ Millions

2,000
1,500
1,000
500
0
1999 2000 2001 2002 2003 2004 2005

Source: Dataquest (May, 2001)

Storage and Bandwidth Price Declines Fuel Storage Demand


As the price of storage and communication bandwidth declines, we believe the
demand for data replication initiatives, such as remote disaster recovery facilities,
could accelerate. Current Dataquest forecasts shown in Figure 19 below, are
projecting external hardware RAID storage prices to decline from $0.15 per
megabyte (MB)/per array unit in 2000 to roughly $0.01 per MB/array unit in 2005,
representing a negative 43% CAGR (2000–05).

33
The SAN Book III – October 22, 2001

Figure 19. Price Trends of External Hardware RAID Storage

0.50

0.40

$/MB/Array Unit
0.30

0.20
-43% CAGR 2000-2005
0.10

0.00
1998 1999 2000 2001E 2002E 2003E 2004E 2005E

Source: Dataquest (July, 2001)

Bandwidth, which provides the communication link for synchronized data


replication, is also facing pricing pressure. A one-year contract that began on July 1
of this year for OC3 bandwidth between New York and Los Angeles now trades for
approximately 31% of its January price. Because of cheaper storage and bandwidth,
data replication should earn higher ROIs, ultimately driving the growth of the
market.

High ROIs in Storage Technology — We Give Numbers


We believe the growing We believe the high financial return earned from new storage technologies is fueling
financial return from
demand for storage. Innovative storage network architectures, such as SANs, and
storage technology is
fueling demand for storage resource management (SRM) software provide significant business and
storage technology. financial advantages and typically generate a high ROI (return on investment) for
the prudent user. The best part: The innovations keep coming — faster, bigger,
cheaper, and better.
Below we highlight the cost savings and benefits from a SAN as seen in our
Citigroup case study.
Factors Driving SAN ROI:
After lengthy conversations with end users and our own research, we believe SANs
have a high return on investment and are an unstoppable trend.
1 Reduced data center floor space requirements: SANs’ increased connectivity
allows organizations to consolidate servers and storage subsystems and typically
results in cost savings on real estate (fewer servers require less data center space).
Citigroup’s direct-attached storage (DAS) environment requires more floor
space to house its decentralized server and storage hardware than in the SAN
environment. In the DAS environment, floor space costs $4.00/GB/month versus
$0.50/GB/month in the SAN. Annually this results in estimated savings of $1.3
million (for the SANs 32 terabytes of storage).

34
The SAN Book III – October 22, 2001

2 Higher storage capacity utilization: In the traditional direct-attached storage


(DAS) architecture, each server has its own dedicated physical storage, creating
individual storage “islands,” where unused capacity cannot be easily moved.
Conversely, a SAN’s centralized architecture has any-to-any connectivity, where
any server can communicate with any storage subsystem in the network. This
architecture allows for greater utilization of storage assets as unused capacity can
be moved to fill storage demand. As an organization increases its utilization rate,
it can delay its storage purchases and generate cost savings (see Figures 16 and
17).
Citigroup utilizes anywhere from 50% to 60% of its storage capacity in its DAS
environment. In the SAN environment, the IT (information technology) group
postpones storage purchases until capacity utilization reaches 90%. This has
resulted in a “storage on demand” ordering process and cost savings because of
delayed purchases.
3 Simplified administration and management: The reduced hardware
requirements attributed to a SAN result in lower management and administration
costs and IT staff can typically manage 5x–10x as much data per person, leading
to reduced labor costs.
Citigroup can manage 10x more storage per person in the SAN (10 TB) than in
the DAS (1 TB) environment. As a result, IT storage managers are able to focus
on strategic issues rather than just storage maintenance. Further, the SAN has
resulted in a more simplified storage procurement process. The order and
installment process can be done in a week in the SAN environment, an
improvement from what used to take six weeks.
4 High system availability: The SAN provides for greater system reliability and
performance and reduces system downtime and lost revenues. For businesses in
intensely competitive environments, an unplanned outage in mission-critical
applications could result in significant loss of revenues, and lower customer
retention and loyalty.
Research indicates that a large financial organization, like Citigroup, has system
downtime costs of up to $6.4 million an hour. Citigroup has noticed a significant
improvement in reliability and downtime in its SAN environment compared to the
DAS. This is due to the SAN’s investment in the core infrastructure whereas the
DAS investment is distributed and is not as robust. According to Citigroup IT
managers, there are probably 10x–20x more outages in a DAS than in a SAN.
5 Reduced hardware requirements: A SANs any-to-any connectivity enables
organizations to consolidate their server hardware and as a result, reduce capital
expenditures and improve corporate cash flow.
In the centralized SAN environment, Citigroup estimates it was able to
consolidate servers and reduce hardware (i.e., servers, subsystems) purchases on
an annual basis.

35
The SAN Book III – October 22, 2001

Storage Technology: Faster, Bigger, Cheaper, Better


We expect storage technology will continue to generate high ROIs because it just
keeps improving. Below we provide a preview of new technologies in the storage
market (a more detailed analysis is provided later in this book):
➤ Faster: This year’s fibre channel switches are expected to transmit data from
rates of 1 Gbps (gigabits per second) to 2 Gbps.
➤ Bigger: The highest port count is currently 128 ports, but is expected to go to
256 by the end of the year. This reduces the need to use extra ports to link two
smaller switches together to get the same usable port count; thereby reducing the
overall cost per port used.
➤ Cheaper: Prices per port have dropped as competition increases and
technology improves.
➤ Better: New features like security, management software, and common
standards are making switches easier to use, especially with ones from different
manufacturers, as well as making them more powerful.
Storage Software
Using effective storage Finally, we believe software advancements are fueling demand for storage
management software technology. One of the most significant developments in this area is Storage
can reduce people
management costs by
Resource Management (SRM) software, which enables IT managers to work with
4x–7x. different systems using one common interface, making the task of managing systems
from various manufacturers easier. The most compelling argument for purchasing
storage management software is that for every $1 spent on storage hardware, it takes
$4–$7 to manage it per year.

Information Growth Is Driving Storage Demand


There are three areas that are the key drivers of information growth and support the
projected growth in demand for storage:
➤ mission-critical applications
➤ the Internet (including e-mail and e-commerce)
➤ emerging killer applications
Worldwide RAID storage, in terabytes, is expected to reach 6.7 million terabytes in
2005 in order to meet the demands of this data growth, representing an 87% CAGR
(2000–05), an improvement from the 81% CAGR for the period 1998–2000.

36
The SAN Book III – October 22, 2001

Figure 20. Worldwide Forecast for RAID Storage (in Terabytes Shipped)

Terabytes
shipped
7,000,000
6,000,000 CAGR 83% (2000-2005)
5,000,000
4,000,000
3,000,000
2,000,000 CAGR 78% (1998-2000)

1,000,000
-
1998 1999 2000 2001E 2002E 2003E 2004E 2005E

Source: Dataquest (July, 2001)

Information Growth from Mission-Critical Applications


The growing reliance on Applications today are much more data intensive than yesterday. Specifically, three
mission-critical
mission-critical applications that are generating significant amounts of data are:
applications, such as
ERPs, is fueling 1 ERPs (Enterprise Resource Planning),
information growth and
demand for storage. 2 CRMs (Customer Relationship Management), and
3 SCMs (Supply Chain Management)
These applications are growing in popularity. One key benefit lies in their ability to
leverage the Internet to make businesses run more efficiently without the additional
expense of more headcount; resulting in increased shareholder value. We believe
the growing organizational reliance on these applications should fuel the demand for
storage in the future.

Enterprise Resource Planning: Projected 12% CAGR


These software applications help procure indirect goods (ones that are not direct
inputs) such as capital equipment and office supplies. Due to their enormous
organizational benefits, ERP applications sales are expected to grow to $46.1 billion
in 2005 from $21.5 billion in 2000, representing a 16% CAGR. As can be seen in
Figure 21 below, ERP applications will continue to grow faster than the overall
business applications market, which has a projected 12% CAGR (2000–05)

37
The SAN Book III – October 22, 2001

Figure 21. Revenue Growth of ERP and Business Applications

140,000
120,000

Revenue ($M)
100,000
80,000
60,000 Bus App: 12% CAGR (2000-2005)
40,000
20,000
ERP: 16% CAGR (2000-2005)
-
1998 1999 2000 2001E 2002E 2003E 2004E 2005E

Business Application Revenue ERP Enterprise-Suite Application Revenue

Source: IDC (February, 2001)

Customer Relationship Management


These software applications improve the sales process. They include such functions
as sales force automation (SFA), customer support, and partner relationship
management. SFA is a powerful organizer that improves productivity by providing
functions such as contact databases, prioritization of accounts, pipeline management,
and forecasting.

Supply Chain Management


SCM software applications provide companies with a high-level view of their
trading partners’ activities. For example, if a manufacturer is planning to increase
production, its suppliers will be notified via SCM software, thereby allowing them
to manage their own production levels more efficiently, avoiding expensive
inventory costs.

Data Warehousing
The growth of data warehousing software, which combines different databases into
one, makes enterprise applications even more powerful, further driving the demand
for storage. Data warehousing provides organizations with the tools to create order
and make sense out of the chaos that can arise out of accumulated information.
Figure 22 details the projected growth of data warehousing software: to $17.0
billion in 2004 from $6.8 billion in 2000, representing a 26.1% CAGR.

38
The SAN Book III – October 22, 2001

Figure 22. Worldwide Data Warehouse Software Revenue Growth Projections

($M)
18,000
16,000
14,000
12,000
26.1% CAGR (1999-2004)
10,000
8,000
6,000
4,000
2,000
0
1999 2000 2001E 2002E 2003E 2004E
Source: IDC (December, 2000)

The Internet 2000 and Beyond


Internet imaging and e- We believe the Internet is driving demand for storage because of the growing
commerce data number of users, their increased reliance on e-mail, and the projected growth of
collection are driving the
demand for advanced
worldwide e-commerce.
data storage capabilities.
E-Mail: Ten Billion Messages per Day in 2000, and Growing
We believe E-mail has a tremendous effect on demand for storage in today’s data
center. In conversations with Citigroup IT executives, e-mail storage is growing
rapidly because “it has become a formal way of doing business and every other e-
mail has some form of attachment.”
According to IDC, on an average day in 2000, ten billion e-mail messages were sent
worldwide. IDC estimates this will grow to 35 billion in 2005, representing a 29%
CAGR.

39
The SAN Book III – October 22, 2001

Figure 23. E-mail Messages Sent on an Average Day

(Billions)
40
35
30
25
Worldwide
20
15
10
North America
5
0
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Worldwide North America

Source: IDC (September, 2000)

Consistent with the Citigroup experience, more users today rely on e-mail to send
content-rich audio and video data, increasing the average size of e-mail and the
storage capacity needed to manage that information. As depicted in the figure
below, it is evident that these new applications are going to require sizable databases
fueling the demand for more storage capacity.
Figure 24. Emerging Application Growth Projections

100
Avg. Application Size (TB)

E-Medicine

50
E-Mail
Data Warehouse
Electronic Security

Video Mail

0
2000 2005 2010

Source: Horizon Information Strategies and StorageTek (1999)

E-Commerce: Projected Five-Year CAGR of 70%


A growing e-commerce business requires an IT infrastructure that has high
availability, and non-disruptive backup/restore processes. Today in a world where
the customer does not sleep, 24x7 system accessibility is critical. Storage software

40
The SAN Book III – October 22, 2001

provides solutions to these needs, through its marquee functions such as data backup
and data recovery. We believe, therefore, that e-commerce growth could drive the
demand for storage in the future.
Ease of accessibility to the Internet in the form of paid Internet access providers like
America Online (AOL) and Internet Service Providers (ISPs) like AltaVista and
NetZero has introduced an increasingly higher number of people to the Internet year
after year. According to May 2001 IDC reports, 370 million devices were used to
access the Web in 2000; a number that is expected to reach 1.3 billion by 2005 —
reflecting a 28% CAGR. E-commerce spending, as shown in Figure 25, is projected
to exceed $5.03 trillion by 2005 up from $354 million in 2000 — a CAGR of
roughly 70%.
Figure 25. Emerging Application Growth Projections

1,400 $6,000

1,200 $5,000
Number of Internet Devices
1,000
(Millions) Left Axis $4,000
28% CAGR 2000-2005

$ Billions
800
Millions

$3,000
600
$2,000
400
E-Commerce Spending
($Billions) Right Axis $1,000
200 70% CAGR 2000-2005

0 $0
2000 2001 2002 2003 2004 2005
Source: IDC (April, 2001)

Emerging Killer Applications


Beyond traditional Beyond traditional storage needs, drivers for storage capacity in the future could
storage needs, we
include emerging “killer applications,” such as:
believe “killer
applications” could ➤ digital photography (see Figure 6)
exponentially drive
future storage demand. ➤ video streaming
➤ set-top boxes, and personal TVs. For example, one minute of telephone-quality
voice is equivalent to 0.36 MB, and one minute of CD-quality sound is
equivalent to 10.6 MB of data.
➤ video conferencing and voice (voice-to-text, text-to-voice, e-mail-to-voice,
voice-to-e-mail, and language translation)

41
The SAN Book III – October 22, 2001

➤ MP3 and MPEG-2 technology


Below we highlight one of the more significant “killer applications” that we believe
could drive future demand for storage:
Enterprise Video Streaming: The North American Market Is
Expected to Grow to $2.8 billion in 2005
Although the adoption of video streaming is still emerging, we believe this
technology is a killer application that could provide for potentially compelling
opportunities in the storage world. As seen in our discussion with the San Francisco
Giants, on-demand streaming video requires a tremendous amount of storage
capacity to enable fast and reliable access to the archived digital data. In North
America, the enterprise video streaming market is expected to grow to $2.8 billion in
2005 from $140 million in 2000, representing a 95% CAGR, according to Jupiter
Media Metrix.
Figure 26. North American Enterprise Video Streaming Market Opportunity

$3.0 $2.8

$2.5 $2.3
(In $billions)

$2.0
95% CAGR (2000-2005)
$1.5 $1.2

$1.0
$0.6
$0.5 $0.3
$0.1
$0.0
2000 2001 2002 2003 2004 2005
Source: Jupiter Enterprise Streaming Model (March, 2001)

Today organizations are using video streaming to improve communication both


internally and externally for such areas as:
➤ executive addresses (earnings releases and annual meetings)
➤ employee education and sales training
➤ new product launches and marketing
➤ B2B collaboration

Video Streaming Growth Drivers


We expect growth in video streaming to be driven by:
➤ returns on investment generated from online training tools that use video
streaming to effectively communicate with employees. Organizations create
corporate value through less expensive (no travel) and more frequent training
programs;

42
The SAN Book III – October 22, 2001

➤ increasing bandwidth penetration by corporations and their employees brought


about by the declining price and increasing availability of bandwidth (as shown
in Figure 27).
Figure 27. Corporate Bandwidth Penetration
(in millions)
( )

% of employees w/broadband
$80 100%

80%
$60
60%
$40
40%
$20
20%

$0 0%
1999 2000 2001 2002 2003 2004 2005

Broadband Dial-up % of Employees with Broadband

Source: Jupiter Internet Access Model (December, 2000)

Storage Opportunity in Streaming Video


Content distribution networks (CDNs) typically use caching for handling streaming
video data in on-demand and live environments. Companies like Network
Appliance offer both software and caching equipment to optimize the performance
of the networks handling streaming video data. In Figure 28, Jupiter Media Matrix
projects the storage opportunity within the streaming video market. Intranet caching
and storage/hosting within streaming video are projected to reach $522 million and
$366 million respectively in 2005, representing respective CAGRs of 109.3% and
71% (2000–05).

43
The SAN Book III – October 22, 2001

Figure 28. Storage Opportunity in Streaming Video

$1,000
Internet Caching/Multicasting CAGR: 109.3% (2000-2005)
$800
Storage/Hosting CAGR: 71% (2000-2005)

(in millions)
$600

$400

$200

$0
2000E 2001E 2002E 2003E 2004E 2005E

Internet Caching/Multicasting Storage/Hosting

Source: Jupiter Enterprise Streaming Model (March, 2001)

Penetration Rates of “Killer Applications”


Figure 29 illustrates the penetration rates of previous “killer applications” to
demonstrate the increasing pace of adoption as consumers become more comfortable
utilizing new technologies.
Figure 29. Previous “Killer Application” Adoption Rates

120%

100%

80%
% of US Households

VCRs

60%

Telephones

40%
PCs

Radio Television Cable TV


20%
Online

0%
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990

Source: FCC, RAB, NAB, NCTA, EIA, IDC, and Salomon Smith Barney

44
The SAN Book III – October 22, 2001

Data Sheet

Size Terminology
As capacity of subsystems grows more each day, the terminology grows more
exotic. Here is a quick explanation of terms you may hear in the future.
Figure 30. Size Terminology: From Byte to Yottabyte

Term Size Example


Byte 8 bits A single character
Kilobyte 1,000 bytes A very short story
Megabyte 1,000,000 bytes A small novel
th
Gigabyte 1,000,000,000 bytes Beethoven’s 5 Symphony
Terabyte 1,000 gigabytes All the X-ray films in a large technological hospital
Petabyte 1,000 terabytes Half the contents of all US academic research
libraries
Exabyte 1,000 petabytes 5 exabytes - all the words people have ever spoken.
Zettabyte 1,000 exabytes As much information as there are grains of sand on
all the world’s beaches
Yottabyte 1,000 zettabytes As much information as there are atoms in 7,000
human bodies
Source: EMC and Salomon Smith Barney

Many forms of traditional media are being converted to digital form so they can be
stored and copied onto computer storage systems. Figure 31 describes the storage
requirements needed for various media.
Figure 31. Sizing for Storage Applications

Source: Horison, Inc.

High Availability Metrics


Availability is an important criterion when talking about mission-critical data.
Manufacturers are striving to improve the reliability of their data. Each minute of
downtime translates into lost productivity for the organization. As the importance of
information grows, this becomes an increasing concern.

45
The SAN Book III – October 22, 2001

Figure 32. High Availability Metrics

Source: Horison, Inc.

Data Is Mission Critical


Just how critical is information? As a company’s reliance on mission-critical
applications grows, so does the value of the information residing within the storage
infrastructure. Downtime costs become exorbitant, especially for e-commerce
companies requiring 24x7 availability to their websites.
Figure 33. Downtime Costs

Application Downtime Costs


Financial Brokerage $6,450,000 / hr
Credit Card Authorization $2,600,000 / hr
Home Shopping $113,000 / hr
Catalog Sales $90,000 / hr
Airline Reservations $90,000 / hr
Tele-Ticket Sales $69,000 / hr
Package Shipping $28,000 / hr
Source: Fibre Channel Industry Association

Figure 34. Downtime Costs — Companies

Application Downtime Costs


Dell Computer $35M / day
Intel Corporation $33M / day
Cisco Systems $30M / day
Amazon.com $4.5M / day
Yahoo.com $1.6M / day
Source: Forrester Research USA

46
The SAN Book III – October 22, 2001

SAN and NAS

What is a SAN?
What is NAS?
SAN and NAS Convergence
What is Server Clustering?

47
The SAN Book III – October 22, 2001

3333333333333333333CQ
What Is a SAN?
➤ A SAN (Storage Area Network) is a storage networking architecture
that enables the more efficient use of storage capacity by decoupling
server and storage connectivity with a dedicated storage network.

What Is a SAN?
A SAN (Storage Area A Storage Area Network (SAN) is a storage networking architecture that enables the
Network) is a more more efficient use of storage assets by decoupling server and storage capacities with
robust and reliable data
architecture.
a dedicated network. Much like a Local Area Network (LAN), a SAN enables
greater scalability and reliability for computer storage resources, making them less
expensive to implement and easier to manage. Traditional storage subsystems attach
directly to the back-end of servers (called Direct Attached Storage, or DAS). In a
SAN, a dedicated storage network is inserted between servers and various storage
subsystems. The storage network is generally composed of a number of switches
which interconnect servers and storage.
Figure 35 below is an example of a traditional DAS architecture.
Figure 35. Traditional DAS (Direct Attached Storage) Architecture
Tape
Library
Key Considerations
1)High cost of management
Client 1
2)Full time data availability
Server 1
often requires traversing the
Subsystem 1 LAN
Client 2
LAN 3)Does not have fast data
transmission when a server
Server 2
has reached its capacity
Client 3 Subsystem 2 4)Does not provide any-to-
any connectivity
5)Scalable at the cost of
Client 4 establishing unused capacity
Server 3 Subsystem 3

Source: Salomon Smith Barney

Figure 36 below is an example of a SAN architecture.

48
The SAN Book III – October 22, 2001

Figure 36. SAN Architecture

Key Advantages
Tape •Reduces overall cost of
Library
storage management
•High availability
Client 1
•Any-to-any connectivity
Server 1 •Frees server capacity
•Reduces LAN congestion
Subsystem 1
Client 2 •Storage not dedicated to
LAN SAN one server
•Increases data transfer
Server 2
Client 3 rates
Subsystem 2
•Enables serverless &
LANless backup
Client 4 •Heterogeneous
Server 3
connectivity (typically
Subsystem 3
between UNIX & NT)
•File sharing

Source: Salomon Smith Barney

Currently, in a SAN, a variety of servers connect to various storage devices via a


Fibre Channel cable, which can be optical or copper. The storage devices form a
pool and are externalized from individual servers, allowing the storage pool to be
shared among multiple servers. This avoids placing additional I/O demand on
existing network data paths and negatively affecting server performance (also
referred to as CPU cycles).
The SAN architecture Through a network of storage devices in a SAN, large numbers of users can
offloads the bandwidth simultaneously access various storage subsystems through multiple paths. This
used by data transfers
between storage
architecture offloads the bandwidth used by data transfers between storage systems
systems from the LAN from the LAN and servers. SAN storage devices can be disk subsystems, tape
and servers. libraries, optical drives, or networking equipment. In a SAN, the storage subsystems
are not dedicated to specific servers, as they are in traditional DAS architectures.
SANs differ from traditional DAS in that:
➤ The cost of management goes down as storage subsystems become centrally
located and easier to manage.
➤ Redundancies are created; in other words, if Server 1 fails in a SAN architecture,
the storage on Subsystem 1 can still be accessed via Server 2. Notice that SANs
can do this without increasing LAN traffic.
➤ Backup or restore (disk-to-tape or tape-to-disk or disk-to-disk) can be done
without involvement of servers and at higher speeds. Notice that SANs can do
this without increasing LAN traffic.
➤ File sharing can occur between servers due to any-to-any connectivity.
➤ SANs scale very efficiently as a result of being able to share storage. Excess
capacity can be allocated where needed in a SAN, instead of only being

49
The SAN Book III – October 22, 2001

available to the server it’s directly attached to, as is the case with a traditional
DAS. In a traditional architecture, excess capacity often goes unused.

Robust SAN Projections


SANs are projected to be SANs are projected to be one of the fastest growing segments of storage. Dataquest
the fastest-growing projects that SAN-attached storage (including external storage and NAS) will grow
segments of storage.
to $41.3 billion in 2005 from $6.3 billion in 2000, representing a 45.7% CAGR.
Dataquest projects that SAN networking (including routers, hubs, switches,
integrated switches, Directors, and HBAs) will grow to $16.9 billion in 2005 from
$1.3 billion in 2000, representing a 67% CAGR. Dataquest’s SAN software
projections (which we include as SAN/Device Administration, SRM, Virtualization
and Backup/Recovery Network) are projected to grow to $8.9 billion in 2005 from
$2.2 billion in 2000, representing a 33% CAGR. These combined segments
represent a 47% CAGR (2000–05).
Figure 37 below illustrates the growth of SANs.
Figure 37. SAN Market Projections
Revenue in millions

$70,000

$60,000 47.0% CAGR


$50,000

$40,000

$30,000

$20,000

$10,000

$0
1998 1999 2000 2001 2002 2003 2004 2005

SAN Storage SAN Networking SAN Software

Source: Dataquest (August, 2001)

SAN growth is expected to exceed that of DAS (which includes JBOD and external
storage not attached to a SAN) and NAS in absolute dollars. Figure 38 below
approximates relative growth of the architectures based on RAID-based disk storage.

50
The SAN Book III – October 22, 2001

Figure 38. RAID-Based Disk Storage in SAN, NAS, and DAS


Revenue in millions

$35,000

$30,000

$25,000
DAS
$20,000

$15,000

$10,000
SAN
$5,000
NAS
$0
1998 1999 2000 2001 2002 2003 2004 2005

Source: Dataquest, 2001

Have No Fear, SANs Are Here


A SAN architecture A SAN architecture improves upon a DAS architecture in several ways including
improves upon a DAS aggregated tape backup, increased capacity utilization, better utilization of high-end
architecture in many
ways.
subsystems through fan-in configuration, centralized location, more storage
managed per person, greater scalability, and higher availability (see Figure 39). By
understanding how these benefits can translate into positive returns for companies,
an investor can better understand the ways in which SANs will progress and which
companies should benefit.
Figure 39 below outlines the reasons for SAN deployment in chronological order of
how we expect they will be utilized.
Figure 39. Comparison of DAS versus SAN

Available
DAS SAN Today in SAN?
1. Aggregate Tape Backup No Yes Yes
2. Capacity Utilization ~50% ~85% Yes
3. Fan-In No Yes Yes
4. Centralized Location No Yes Yes
5. GB Per Person Managed <100 >500 Yes
6. Scalable architecture No Yes Yes
7. High Availability No Yes No
Source: Salomon Smith Barney

1. Aggregate Tape Backup


One of the most tangible, real world benefits of SAN adoption we hear about often
from IT managers is streamlining the tape backup process. Backing up data to tape
is a less expensive alternative to backing up to disk. Although it is slower, the cost
of tape per megabyte is often less than $0.01, or about one-fiftieth of the cost of a

51
The SAN Book III – October 22, 2001

megabyte of disk storage. Hence tape is often used to store data that is accessed less
frequently or for archival purposes.
In a DAS environment, each server and storage subsystem must have its own
separate tape device or drive (see Figure 40 below). The main disadvantage to this
is the management and collection process. In this layout, each tape drive could be
located far away from the other. (One IT manager jokingly suggested his employees
should wear roller skates so they could collect all of the tapes located throughout the
enterprise.) Since a single-tape drive can only store around 100 GB, tapes often
have to be replaced on a daily basis and filed into a Dewey Decimal type library.
Figure 40. Tape Backup Using DAS

Client 1
Tape
Server 1 Subsystem 1 Drive 1

Client 2
LAN
Tape
Server 2 Subsystem 2 Drive 2
Client 3

Client 4 Tape
Subsystem 3
Server 3 Drive 3

Source: Salomon Smith Barney

Another alternative is to back up over the LAN, which we discuss in a later section
titled “The Future of SANs.”
In a SAN environment In a SAN environment, the tape backup process is improved by being centrally
the tape backup process located and thus easier to manage. The collection of subsystems on the network can
is improved by being
centrally located and
back-up their data to an individual tape library instead of to geographically-
thus easier to manage. dispersed, single-tape drives. Tape libraries are enterprise-level automated devices
that hold multiple tapes and thus more total storage. For example, StorageTek’s
PowderHorn 9310 tape library can store up to 360 TB of data (versus 100 GB in a
tape drive). Since it can hold more data than a single-tape drive, the tapes do not
have to be collected as frequently.

52
The SAN Book III – October 22, 2001

Figure 41. Tape Backup in a SAN

Client 1

Server 2 Subsystem 1

Client 2
LAN SAN
Server 2 Subsystem 2 Tape
Library
Client 3

Client 4
Server 2 Subsystem 3

Source: Salomon Smith Barney

2. Capacity Utilization
Another benefit to SANs is higher capacity utilization. In a DAS environment, each
server typically has its own dedicated subsystem. In talking with IT managers, we
have found 50% to be a typical capacity utilization rate (percentage of total available
storage being used) in these environments.
We have found 85% to In a SAN environment, multiple servers can store data on a single larger subsystem
be a typical capacity (see Figure 42 below). In a SAN environment, we have found 85% to be a typical
utilization rate in a SAN
environment versus 50%
capacity utilization rate. IT managers often allow higher utilization rates for
in a DAS environment. subsystems in a SAN environment because it is easier to monitor a single storage
unit reaching its maximum limit versus many units in a distributed DAS
environment.
Figure 42. Better Capacity Utilization in a SAN

DAS SAN
8 small subsystems, 1 large subsystem,
50% utilization each 85% utilization
8 TB of storage used 8.5 TB of storage used
16 TB of storage capacity 10 TB of storage capacity

Server 1 Subsystem 1 Server 5 Subsystem 5


Server 1

Subsystem 2 Server 6 Subsystem 6 SAN


Server 2
Server2

Subsystem 1
Server 3 Subsystem 3 Server 7 Subsystem 7

Server 3

Server 4 Subsystem 4 Server 8 Subsystem 8

Source: Salomon Smith Barney

53
The SAN Book III – October 22, 2001

3. Fan-in Scale Leads to Higher ROI


Economies of scale In a DAS environment, for each server there is one subsystem. Hence, many smaller
make the larger subsystems must be purchased to accommodate this requirement. In a SAN
subsystem less
expensive while
environment, only a few subsystems are needed to accommodate many servers (see
providing greater Figure 43 below). We refer to the many (server)-to-few (subsystems) relationship as
functionality. “fan-in” because of the “V” shape it resembles. For the same amount of money it
would take to purchase the many small subsystems needed in a DAS environment, a
company could choose to purchase fewer large subsystems, which typically have
more features than their smaller counterparts and at a lower cost per megabyte.
Economies of scale make the larger subsystem less expensive while providing
greater functionality. Hence, by using a fan-in layout, an IT manager can get more
for each dollar spent.
Figure 43. Many-to-Few “Fan-in” Saves Money or Utilizes Higher-end Subsystems

DAS SAN
8 servers + 8 servers +
8 smaller, mid-range subsystems 3 larger, enterprise subsystems

Server 1

Server 1 Subsystem 1 Server 5 Subsystem 5


Server 2
Subsystem 1

Server 3
SAN
Server 2 Subsystem 2 Server 6 Subsystem 6 Server 4

Server 5
Subsystem 2

Server 6
Server 3 Subsystem 3 Server 7 Subsystem 7

Server 7

Subsystem 3
Server 8
Server 4 Subsystem 4 Server 8 Subsystem 8

Source: Salomon Smith Barney

4. Centralized Location
Another benefit to a SAN is the centralized location. In a DAS environment, servers
are often located in multiple locations within an enterprise. Since DAS typically
uses interconnects with a limited distance capability (e.g., Ultra 160 SCSI has a limit
of 12 meters), the storage must be located close to the server. Hence, storage
subsystems must be located close to their respective server counterparts and cannot
be aggregated.
By locating all of the In a SAN environment, the storage can be located further away from the server since
storage in one location it is networked. For example, using Fibre Channel, storage can be located 10 km
using large, dense
subsystems versus
away. Additionally, by using channel extenders (devices that extend the length at
smaller ones, less floor which data can be transmitted), storage can be located as far as 200 km. As a result,
space is taken up per all of an enterprise’s storage can be located in one central location. By locating all
megabyte. of the storage in one location using large, dense subsystems versus smaller ones, less
floor space is taken up per megabyte. Additional cost saving can be realized by

54
The SAN Book III – October 22, 2001

spreading the fixed costs of real estate (i.e., air conditioning, lighting, security) over
more gigabytes.
5. Low Cost of Management: More GBs Per Person
A SAN architecture A SAN architecture maximizes the efficient use of each IT administrator through its
maximizes the efficient user-friendly feature/functions and its design to be physically centrally located.
use of each IT
administrator.
StorageTek recently researched and measured the benefits of SANs and found that:
➤ In a distributed storage environment, each IT administrator can manage
approximately 30–100 GB of data, generating people costs representing an
estimated 55% of the overall storage budget.
➤ In a centralized storage environment, each IT administrator can manage 500 or
more GBs of data, lowering people costs to an estimated 15% of the overall
storage budget.
Figure 44. Cost Advantage of Centralized Storage

$/MB/MO
3 Cost without storage
management software
Disk Storage

2
ost
s onC
Per
1 Cost with storage
management software

0
1978 1984 1995 2000

Source: IBM, Strategic Research Corp.

6. Scalability
A SAN can allocate SANs scale very efficiently, which is not to say that traditional DAS does not scale.
excess capacity where The advantage of a SAN is that it can allocate excess capacity where needed. In
needed.
Figure 45 below, in a traditional DAS architecture, excess storage capacity on
Subsystem 1 would go unused if Server 1 did not need it. In a SAN architecture,
excess storage capacity on Subsystem 1 could be allocated to Server 2 if needed.

55
The SAN Book III – October 22, 2001

Figure 45. SAN Is More Scalable

DAS SAN
Subsystem 1: Utilization 10% Subsystem 1: Utilization 55%
Subsystem 2: Utilization 100% Subsystem 2: Utilization 55%
Subsystem 2 is full Neither subsystem is full

Subsystem 1 Server 1 Subsystem 1


Server 1

SAN

Server 2 Subsystem 2

Subsystem 2
Server 2

Source: Salomon Smith Barney

7. Future of SANs: Any-to-Any Connectivity


One of SAN’s most In our view, one of SAN’s most promising features is any-to-any connectivity, the
promising features is ability for any server to talk to any subsystem. Any-to-any connectivity enables new
any-to-any connectivity,
the ability for any server
applications that create a high availability architecture that better utilizes servers,
to talk to any subsystem. network and storage subsystems resources.
The same demands in a SAN architecture could be handled by using Server 1 and
Server 2 simultaneously to access data on Subsystem 1, also known as load
balancing (see Figure 46 below). In a SAN architecture, in theory, any server can
access any storage subsystem. This is any-to-any connectivity.
Figure 46. Any-to-Any Connectivity

Client 1
Server 1 Subsystem 1

Client 2
LAN SAN
Server 2 Subsystem 2
Client 3

Client 4
Server 3 Subsystem 3

Source: Salomon Smith Barney

56
The SAN Book III – October 22, 2001

Automatic Failover Provides Greater Data Availability


Through any-to-any Through any-to-any connectivity, SANs help increase the ease and reliability of data
connectivity, SANs help recovery through architecting multiple paths.
increase the ease and
reliability of data In Figure 47 below, we illustrate how a traditional DAS backup traverses computing
recovery through devices.
architecting multiple
paths. Note: The thick lines represent the data path in the figures that follow.
Figure 47. Traditional DAS (Direct Attached Storage) Backup
Tape
Library
DAS
(Traditional Storage Backup)
Client 1
Server 1 •In a traditional DAS
architecture, backup traffic
Subsystem 1
must traverse an already
Client 2
LAN congested LAN (often at
night or during non-business
Server 2
hours).
Client 3 Subsystem 2
•DAS backups involve
application servers to run the
backup application software.
Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

DAS backups require the involvement of application servers and are, therefore,
interrupted when servers fail (see Figure 48 below).
Figure 48. Traditional DAS (Direct Attached Storage) Backup Stops when a Server Fails
Tape
Library

Client 1 DAS
Server 1
(Traditional Storage Backup)
Subsystem 1
•In a tradit ional DA S
Client 2
LAN architecture, server and
network availability is vital.
Server 2 •DAS backups cannot be
Client 3 Subsystem 2
completed when a server
fails.

Client 4
Server 3 Subsystem 3

Source: Salomon Smith Barney

57
The SAN Book III – October 22, 2001

The SAN architecture The SAN architecture creates redundancies that result in the High Availability (HA)
creates redundancies of data through any-to-any connectivity. For example, if Server 1 fails in a SAN
that result in the High
Availability (HA) of data
architecture, Subsystem 1 can still be accessed from Server 2. Notice that SANs can
through any-to-any do this without increasing LAN traffic.
connectivity.
In a SAN environment, server failures do not interrupt data access. (see Figure 49
below)
Figure 49. SANs Create Multiple Paths to Circumvent Server Failures
Tape
Library

Client 1
Server 1
SAN
•In a SAN, a failed
Subsystem 1 server can be
Client 2
LAN circumvented.

Server 2
Client 3
Subsystem 2

Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

In a SAN environment, storage networking failures do not interrupt data access.


(See Figure 50 below.)
Figure 50. SANs Create Multiple Paths to Circumvent Storage Networking Failures
Tape
Library

Client 1
Server 1
SAN
•In a SAN, a storage
Subsystem 1 networking failure
Client 2
LAN can be circumvented.

Server 2
Client 3
Subsystem 2

Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

In a SAN environment, storage subsystems failures do not interrupt data access to


mirrored data (i.e., an exact duplicate of the data) and data can be stored on multiple
storage subsystems as illustrated in Figure 51 below.

58
The SAN Book III – October 22, 2001

Figure 51. SANs Create Multiple Paths to Circumvent Storage Subsystem Failures
Tape
Library

Client 1
Server 1
SAN
•In a SAN, data can
Subsystem 1
be backed up on
Client 2
LAN multiple storage
subsystems and
Server 2 mirrored data can be
Client 3
Subsystem 2 accessed from
multiple storage
subsystems.
Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

The SAN Brain


Who controls the data Who controls the data traffic in a SAN? The most popular answer we have heard is
traffic in a SAN? The “me.” Everyone wants to control the SAN. If SANs are going to be as great as
most popular answer we
have heard is “me.”
everyone we spoke with thinks they will (and we agree), the most value will be
added by whoever controls the SAN (see Figure 52 below).
Figure 52. Potential SAN Manager

SAN
Manager? Tape
Library

Client 1
Server 1

Subsystem 1
Client 2
LAN SAN

Server 2
Client 3
Subsystem 2

Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

In our view, there are four viable contestants:


1 Application Servers: This is how traditional DAS has been controlled in the
past. This process has to change, however, because application servers are too

59
The SAN Book III – October 22, 2001

expensive to baby-sit data backups, especially since many companies have


already illustrated that less expensive servers (thin servers) can do it just as well.
2 Storage Subsystems: We believe this would be the tail wagging the dog. By
building up the intelligence in storage subsystems, the cost is simply being
migrated from application servers, without really enhancing performance. This
has been a logical first step toward freeing up application server’s capacity.
3 SAN Appliance: We believe this is the most viable alternative: Install a
dedicated appliance that is simple to use in the middle of the SAN and let it be
the traffic cop via in-band or out-of-band management. Many startup companies
are betting their futures on this concept. We think this is a viable solution.
4 Networking Equipment: To many, this seems like a natural development. If a
switch or router, combined with server capabilities, was improved by adding
various software and hardware features, many believe it could be a very effective
SAN manager.
In the end, we believe multiple types of devices and appliances will offer various
functions toward the same goal: efficient data management.

SAN Adoption
We believe adoption Despite recent economic pressure within the industry, we are still firm believers in
rates are going to SAN adoption. We believe adoption rates are going to dramatically increase when
dramatically increase
when the economy
the economy stabilizes and IT budgets return to normal levels. That is not to say
stabilizes and IT budgets that SAN adoptions have stopped in this environment, but they have slowed along
return to normal levels. with everything else. Additionally, the availability of serverless backup and other
SAN “killer applications” should also drive future growth.
Figure 53 below illustrates the increasing adoption rate of SANs. Going forward,
the SAN architecture will represent a larger percentage of the total market based on
RAID-based disk storage growth.

60
The SAN Book III – October 22, 2001

Figure 53. RAID-Based Disk Storage by Architecture

100%

80%
DAS
60%
SAN
40%

20%
NAS
0%
1999 2000 2001 2002 2003 2004 2005

NAS SAN DAS

Source: Salomon Smith Barney

We have heard people talking about how many SANs have actually been installed.
The key premise to understand is how SAN is being defined. Some define it as
Fibre Channel revenues, while others define it as multiple servers (more than two)
being connected to multiple storage devices (more than two). Although the former
is more common, we believe the latter is more accurate.
Technology has historically been an area in which adoption rates can linger. We
estimate “the year of the SAN” will be no different; taking off over the course of the
next several years and experiencing a protracted adoption period.
In the end we expect In the end we expect SANs to move from about 5% penetration today to 90% in the
SANs to move from future. From an account penetration point of view we believe SAN penetration is
about 5% penetration
today to 90% in the
closer to 15%. But note, not all customers who have adopted SANs have rolled it
future. out into 100% of their environment. We believe even existing adopters are only
10%–30% penetrated from a potential addressable port connectivity point of view
(again, on their way to 90%).

61
The SAN Book III – October 22, 2001

What Is NAS?
➤ NAS (Network Attached Storage) is a storage networking architecture
that allows storage subsystems to connect directly onto the existing
computing network, the Local Area Network (LAN).

What Is NAS?
NAS (Network Attached NAS (Network Attached Storage) is a storage networking architecture that allows
Storage) is a storage storage subsystems to connect directly onto the existing computing network, the
networking architecture
that allows storage
Local Area Network (LAN). Traditional storage subsystems apply Direct Attach
subsystems to connect Storage (DAS) to the back end of servers and have used costly application server
directly onto the existing capacity to transfer data from storage devices to clients (desktop computers). This
Local Area Network “application server tax” is avoided in the NAS architecture by attaching a smaller,
(LAN).
less expensive server (referred to as a thin server, file server, or Filer) to a disk
subsystem and directly attaching the combined subsystem (referred to as a NAS
appliance) onto the LAN, thus forgoing the traditional dependence of file system
access on application servers (see Figure 54 below).
Figure 54. Traditional NAS Architecture

Subsystem 1
Key Advantages
•Frees server capacity
•Storage not dedicated to
Client 1
one server
Server 1
Tape •Uses existing TCP/IP
Library
Client 2 networks
LAN •Client can directly access
Subsystem 2
Server 2 data
Client 3
•Heterogeneous
connectivity (between
Client 4 UNIX & NT)
NAS •Simplifies Data
Appliance
Management

Source: Salomon Smith Barney

Through the NAS architecture, storage subsystems gain access to multiple hosts
(servers and desktops), as seen in Figure 54 above. For example, the NAS appliance
can speak to Client 1 or Client 2 without major application server involvement.
Figure 55 below lists major differences in NAS and SAN.

62
The SAN Book III – October 22, 2001

Figure 55. NAS Versus SAN

SAN NAS
Protocol Fibre Channel NFS, CIFS
protocol
Network Fibre Channel Ethernet
Transfer Block File
Embedded File System No Yes
Source: Salomon Smith Barney

A NAS architecture A NAS architecture lowers storage management costs by simplifying the storage
lowers storage process and achieving heterogeneous connectivity, thereby enabling file sharing
management costs by
simplifying the storage
across multiple platforms which allows storage to be centrally managed.
process and achieving Note: Unless stated explicitly, we will refer NAS to mean disk-based NAS, not
heterogeneous
connectivity. optical NAS, nor tape-based NAS.

Key Advantages of NAS Architecture


The NAS architecture offers several benefits, including:
➤ Reliability, Availability, and Serviceability (RAS);
➤ The simple “appliance” design of NAS devices increases reliability vis-à-vis
general-purpose servers, reducing both IT staff and end-user hassles;
➤ NAS storage is not dedicated to one server (in traditional client-server
architectures, if a server goes down, none of the files on an attached storage unit
can be accessed);
➤ Clients can directly access data over a LAN without going through a server,
freeing server capacity for other tasks and thus improving network performance;
➤ NAS can be implemented over existing TCP/IP networks; and
➤ Adding storage capacity does not require taking a general-purpose server down,
eliminating the need for IT personnel to work off-hours to add capacity. Should
a server be down during business hours, Gartner Group determined that the cost
of lost user productivity ranges from $200 to $2,000 per hour.
Manageability
➤ Heterogeneous connectivity between UNIX and NT, enabling file sharing across
multiple platforms (traditional client-server architectures require separate copies
of files for NT and UNIX); and
➤ Simplified data management (for example, in traditional client-server
architectures, if the NT version of a file is changed, the UNIX version then
needs to be updated and vice versa).
Performance
➤ Because the NAS appliance is running a thinner operating system than a
general-purpose server, it can deliver similar or higher performance with less
computing power.

63
The SAN Book III – October 22, 2001

Low Total Cost of Ownership (TCO)


➤ Simple, reliable “appliance” design results in fewer failures and quicker, easier
fixes, which saves on IT staff costs and allows personnel to be deployed to other
meaningful projects;
➤ Simple “appliance” design results in lower upfront cost than is the case for
general-purpose servers; and
➤ Quick, easy installation (typically less than half an hour), which saves on IT
staff costs.

However, There Is No Free Lunch


➤ Lower-end NAS devices are not fault-tolerant;
➤ NAS is scalable at the cost of losing its simplicity and manageability;
➤ NAS increases network traffic on the LAN (unless a separate LAN is created);
and
➤ Data transmission speeds are constrained by the LAN, limiting performance.
➤ While NAS distributes the compute cycles to less expensive file servers, it
creates a greater total compute demand on the application servers since it has to
process Ethernet packets versus SCSI blocks. For a better understanding of file
versus block transfer, please see this report’s “Software RAID” section.

NAS Market Segmentation


NAS appliances span a The NAS market is generally divided into three segments: entry-level, mid-range,
wide range of prices and and high-end. As the devices move up-market, they feature not only more storage
capabilities to meet a
variety of market needs.
capacity but also many of the reliability, availability, serviceability, scalability, and
manageability features found in higher-end general-purpose servers. Examples
include snapshot-based data recovery, remote mirroring, redundant and hot-
swappable components, status reports, failure alerts, diagnostic capability, and
integration with enterprise management tools. Much of this value-add in the higher-
end NAS devices is in the software.
Figure 56 below outlines which companies participate in these segments.

64
The SAN Book III – October 22, 2001

Figure 56. Market Positioning of NAS Vendors

Auspex
High-End Broadband Storage
Compaq
EMC
Hewlett Packard
IBM
Midrange Network Appliance
Network Storage Solutions
Price Procom
3Ware Sun Micro
Hewlett Packard
IBM
Entry Maxtor
Network Storage Solutions
Intel Procom
Maxtor Quantum (Connex)
Nortel
Quantum

Performance/Availability
Source: Salomon Smith Barney

Entry NAS
These are products designed to provide inexpensive, simple file sharing for smaller
workgroups, or as an alternative to portable storage technologies such as Iomega’s
external Jaz. The uses of these products may be permanent for small businesses —
for example, replacing the need for a traditional file server. Examples of products in
this class are Quantum’s Snap Servers and Maxtor’s MaxAttach.

Midrange NAS
These products are designed to have increased performance, scalability, and
reliability from Entry NAS. Pricing typically ranges from $5,000 to $100,000 with
maximum capacities scaling to 2 TB. Examples of products in this class are
Network Appliance’s F85 and Compaq’s TaskSmart N2400.

High-end NAS
Examples of NAS These products are designed for use in data centers that are expected to provide
products in the various multiple terabytes of storage to the network or application servers. The products
segments can be found
in the “NAS
must deliver high availability and scalability, and thus they are characterized as
Subsystems” section. having more than one storage controller; nearly always having redundant
components, multiple processors, and network interface cards; and usually being the
first to adopt new performance-oriented storage technologies. Pricing starts at
$100,000 with capacities scaling to 12 TB in clustered configurations. Some
examples of products in this class are EMC’s Celerra and Network Appliance’s
F800 Series.
Examples of NAS products in the various segments can be found in the “NAS
Subsystems” section.

65
The SAN Book III – October 22, 2001

Market Share and Projections


Dataquest estimates the overall NAS market will grow to $10.3 billion in 2005 from
an estimated $1.5 billion in 2000, representing a 48% CAGR (Figure 57 below).
Figure 57. NAS Revenues and Projections
($ in millions) CAGR
Price Segmentation 1998 1999 2000 2001 2002 2003 2004 2005 2000-2005
Greater than $100,000 $262 $416 $1,033 $1,380 $2,336 $3,312 $4,329 $5,603 40.2%
$25,000 to $100,000 $87 $128 $284 $323 $647 $1,107 $1,734 $2,637 56.1%
$5,000 to $25,000 -- $9 $44 $142 $323 $585 $957 $1,476 101.7%
$2,000 to $5,000 -- $4 $54 $73 $167 $280 $407 $533 58.1%
Less than $2,000 -- $24 $32 $32 $49 $62 $77 $93 23.7%
Total RAID Revenue $349 $580 $1,448 $1,950 $3,521 $5,347 $7,502 $10,343 48.2%
Year-to-Year Growth -- 66.2% 149.7% 34.7% 80.6% 51.8% 40.3% 37.9% --

Source: Dataquest (August, 2001)

While the high-end only While high-end NAS ($100,000 and above) only commanded 6% of total NAS
commanded 6% of total units, it dominates total revenues, representing 71% of total NAS revenues in 2000.
NAS units, it dominates
total revenues,
Conversely, the entry level ($25,000 and below) has approximately 85% of total
representing 71%. units but represented only 9% of total revenues in 2000 (Figure 58 below).
Figure 58. NAS Units and Projections

CAGR
Price Segmentation 1998 1999 2000 2001 2002 2003 2004 2005 2000-2005
Greater than $100,000 1,514 2,285 3,834 4,870 8,959 13,833 20,481 29,312 50.2%
$25,000 to $100,000 1,527 2,337 5,486 6,036 15,364 27,620 46,041 74,543 68.5%
$5,000 to $25,000 -- 1,083 3,852 10,727 21,325 35,636 58,772 91,650 88.5%
$2,000 to $5,000 -- 1,275 18,495 26,575 59,680 104,211 158,566 227,004 65.1%
Less than $2,000 -- 19,690 30,500 33,725 47,706 62,756 78,611 96,521 25.9%
Total RAID Revenue 3,041 26,670 62,167 81,933 153,034 244,056 362,471 519,030 52.9%
Year-to-Year Growth -- 777.1% 133.1% 31.8% 86.8% 59.5% 48.5% 43.2% --

Source: Dataquest (August, 2001)

Network Appliance, EMC, and Quantum, combined, have over 90% of the total
NAS market share. Figure 59 below illustrates NAS market share by revenue in
2000.
Figure 59. Total NAS Market Share by Revenue, 2000

Dell
2.0% Others
7.9%
Quantum
4.0%
Network Appliance
49.8%

EMC
36.3%

Source: Dataquest (April, 2001)

66
The SAN Book III – October 22, 2001

SAN and NAS Convergence


➤ We believe the trend to network storage is bringing focus to a new
“Data-Centric” environment that merges many storage architectures
and technologies.

➤ In our opinion, the merging of SAN and NAS will further drive more
demand for storage networking software and equipment, similar to
the way LANs increased demand for PCs and servers.

Data-Centric Architectures
We believe SAN, NAS, We believe all debates around NAS, SAN, and traditional DAS (Direct Attached
and DAS architecture Storage) should dissipate in favor of one “Data-Centric” architecture. In a Data-Centric
have been merging and
will continue to merge.
architecture, nobody cares what you call it, whether it be SAN, NAS, Fibre Channel,
iSCSI, or Ethernet. The only thing that matters is whether or not it works! Keep in
mind, the whole reason SAN and NAS have emerged is to solve today’s data puzzle,
which is: 1) data is doubling every year, and 2) IT staffing budgets are remaining flat.
In other words, customers need to manage more data, but with no additional resources.
The other important factor to keep in mind is that nobody is going to throw out current
equipment just because newer equipment is better. Simply stated, newer equipment that
is backward-compatible with the current equipment is what customers are most likely
going to buy. This benefits everyone: 1) customers migrate toward better, more
advanced solutions, 2) customers get to reap the benefits of past investments, and 3)
new technologies have the opportunity to thrive through adoption.

Notice that when NAS Is Scaled, It Becomes a SAN


When NAS is scaled it Although the NAS architecture is technically scalable, in so doing, it becomes more
becomes a SAN. complex and loses its original appeal of simplicity. When NAS scales, it also
increases traffic on an already burdened LAN.
Figure 60 below depicts a traditional NAS architecture.

67
The SAN Book III – October 22, 2001

Figure 60. Traditional NAS Architecture

Subsystem 1

Client 1
Server 1
Tape
Library
Client 2
LAN
Subsystem 2
Server 2
Client 3

Client 4

NAS
Appliance

Source: Salomon Smith Barney

To add greater tape backup capacity, another tape library must either be added to the
LAN, to the back-end of an application server, or to the back-end of the NAS (the
latter of which we would call a SAN if it interconnected multiple NAS subsystems).
To add more disk capacity, another NAS appliance must be added to the LAN or to
the back of the NAS.
Adding tape connectivity Figure 61 below illustrates how adding tape connectivity to the back-end of NAS
to the back-end of NAS develops a NAS/SAN hybrid.
develops a NAS/SAN
hybrid. Figure 61. By Scaling NAS, It Becomes a SAN
Tape
Library

Key Considerations
1)Relatively low cost of
Client 1
Server 1
management
Disk
2)Fault tolerant
Array 1 3)Data transmission speeds
Client 2 NAS Disk
LAN Appliance 1 are somewhat constrained
by the LAN
4)Heterogeneous
Client 3 N
SA connectivity
NAS Disk
5)Scalable at the cost of
NAS Tape
Appliance 2 Appliance losing its simplicity
Client 4

Source: Salomon Smith Barney

68
The SAN Book III – October 22, 2001

When scaling a NAS When scaling a NAS architecture, it makes the most sense to us (with today’s
architecture, it makes the technologies) to merge NAS and SAN architectures to achieve the most efficient
most sense to us (with
today’s technologies) to
results. For this reason, we believe SAN and NAS architectures complement one
merge NAS and SAN another.
architectures to achieve
the most efficient Network Appliance FIlers Are Interoperable with Brocade SANs
results. In May, 2001, Network Appliance announced that its filers were interoperable with
Brocade SANs, allowing configurations like the one depicted in Figure 61 above.
By supporting SAN-enabled backups, we believe Network Appliance opens up its
solution to greater market potential. By putting the tape backup solution on a SAN
behind NAS instead of onto the LAN, Network Appliance filers offer reduced
network traffic on the LAN, high availability of the tape appliance, high availability
of its Filers, and greater scalability, thereby creating a more robust computing
environment.

SANs Also Enable Disk Mirroring for NAS


Some NAS companies have even begun using their Fibre Channel back-ends to
provide failover, by creating multiple data paths, and greater reliability, through disk
mirroring. Network Appliance, the leading NAS appliance provider, is one of the
companies that could use its Fibre Channel back-end to scale into data center
environments by creating a SAN back-end to ensure greater reliability.
Figure 62 below depicts an environment using disk mirroring (or replication) for
NAS.
Figure 62. Using a SAN Behind NAS to Provide Failover and Disk Mirroring

Tape
Library 1

Client 1

Server 1
Tape
Client 2 NAS Disk Disk Library 2
LAN Appliance 1 Array 1

Disk
Client 3 N
SA
Replication 1

Disk
NAS Disk Replication 2
Appliance 2
Client 4

Source: Salomon Smith Barney

69
The SAN Book III – October 22, 2001

EMC’s Celerra NAS Server Is a SAN


We believe EMC’s We believe EMC’s Celerra is one of the best ways to service the high-end,
Celerra is one of the best expandable NAS market. EMC has made inroads into the high-end NAS market
ways to service the high-
end, expandable NAS
over the past year resulting in annual NAS revenues of $524 million in 2000.
market. Celerra has gained significant momentum recently, reaching revenues of
approximately $562 million in the first six months of 2001.
Celerra is a more advanced approach to NAS in that it is not a complete NAS
appliance, but a series of NAS servers (thin file servers) without the traditional disk
integration. Celerra attaches a Symmetrix disk subsystem (EMC’s traditional disk
subsystem) on its back-end, potentially forming a type of SAN architecture behind
Celerra (we would refer to this design as a SAN/NAS hybrid). In other words,
Celerra provides a gateway for clients to access files directly over the LAN
(typically Symmetrix disk subsystems are directly attached to application servers,
therefore, forcing clients to traverse application servers in order to retrieve files).
Although the Celerra Although the Celerra architecture loses the simplicity of NAS, it captures all other
architecture loses the features while increasing reliability, scalability, and availability. Figure 63 below
simplicity of NAS, it
captures all other
illustrates how a Celerra-based NAS architecture might look.
features while increasing
Figure 63. Traditional Celerra-Based NAS Architecture
reliability, scalability,
and availability.

Tape
Library

Key Considerations
1)Relatively low cost of
Client 1
Server 1
management
Symmetrix 1
2)Fault tolerant
Client 2
3)Data transmission speeds
LAN Celerra 1
are somewhat constrained
by the LAN
Symmetrix 2
4)Heterogeneous
Client 3
connectivity
5)Scalable at the cost of
losing its simplicity
Client 4 Celerra 2 Symmetrix 3

Source: Salomon Smith Barney

Figure 64 below illustrates how a Celerra-based SAN architecture might look.

70
The SAN Book III – October 22, 2001

Figure 64. Potential Celerra Based SAN Architecture


Tape
Library 1

Key Considerations
1)Relatively low cost of
Client 1
Server 1
management
EMC
2)Fault tolerant
Symmetrix 1 3)Data transmission speeds
Client 2
LAN EMC
Celerra 1 are somewhat constrained
by the LAN
EMC
Symmetrix 2 4)Heterogeneous
Client 3 N
SA connectivity
EMC
5)Scalable at the cost of
Celerra 2 EMC losing its simplicity
Client 4 Traditional Symmetrix 3
NAS
Appliance
Tape
Library 2

Source: Salomon Smith Barney

The Future Could Have It All


Looking into the future Looking into the future we envision a more diverse “Data-Centric” architecture that
we envision a more incorporates and migrates from DAS to SAN while leveraging the benefits of NAS.
diverse “Data-Centric”
architecture that
Figure 65 below illustrates this architecture. Note how it incorporates legacy
incorporates and investments.
migrates from DAS to
SAN while leveraging the Figure 65. Data-Centric Architecture
benefits of NAS.
WAN/MAN/LAN
Tape
DAS Library

Application
Subsystem
Client 1 Server 1
1

Subsystem 2
Client 2
LAN SAN
Application
Server 2
Client 3
Subsystem 3

NAS File Server


NAS Tape
Appliance NAS
Appliance 1 NAS Subsystem 4
Appliance 2 High-end,
SAN/NAS
Traditional, hybrid
low-end NAS
Source: Salomon Smith Barney

71
The SAN Book III – October 22, 2001

What Is Server Clustering?


➤ We believe that server clustering has been one of the least
recognized networking growth opportunities. In our opinion, server
clustering will be a key area of growth for today’s storage software
and networking companies.

What Is a Server Cluster?


We are big believers in Server Clusters are groupings of servers networked together to form clusters. There
clustering and view the are three types of server clusters: 1) failover, 2) parallel processing, and 3) load-
storage software and
networking companies
balancing. We will discuss failover clustering since this is one of the fastest
as the natural winners. growing areas.
In a failover server cluster, servers are grouped together to create High Availability
(HA) sub-networks referred to as a system area network (san, as opposed to a SAN),
which ensures greater application availability (sometimes measured using the nines
of availability or 99.999% uptime). Said simply, failover server clustering allows a
secondary server (or servers) to take over the primary server’s functions in case it
fails. Reliability, availability, and flexibility of the clustering deployment (such as
adding or removing systems or applications from a cluster) are the key features of
clustering.
Note: Workstation and storage subsystems can also be clustered to form system area
networks (sans).
Failover server clustering is gaining momentum as the importance of data and the
penalties for downtime increase. Figure 66 below illustrates a potential cluster
server configuration.
Figure 66. Servers Are Clustered to Provide Greater Reliability

Servers 1 & 2

Server 6 Server 3

Servers 4 & 5

Source: Salomon Smith Barney

Switches could also be used to enable higher node count clusters as depicted in
Figure 67 below.

72
The SAN Book III – October 22, 2001

Figure 67. Server Clustering Using Storage Network Switches

Source: Salomon Smith Barney

Taking things a bit further, we thought it useful to depict a full-blown server-


clustered, SAN-connected computing environment (Figure 68 below).
Figure 68. Scalable Storage Networking Cluster

Tape
Library 1
Client 1

Client 2 Subsystem 1

Subsystem 2
Client 3
LAN

SAN

Subsystem 3
Client 4
Subsystem 4

Client 5
Subsystem 5

Client 6 Subsystem 6

Client 7 Tape
Library 2
Subsystem 7

Source: Salomon Smith Barney

73
The SAN Book III – October 22, 2001

Figure 69 below illustrates another way to unite server clusters and storage
networks, potentially using multiple protocols, including FC-SCSI, FC-IP, and FC-
VI, enabling server-to-server, server-to-storage, and storage-to-storage connectivity.
This is the ultimate system area network.
Figure 69. Designing a Potential Systems Area Network (san)

s
ver
Se r
Clients

Storage
LAN

san

S er
ver
s

Source: Salomon Smith Barney

Clustering Could Be Huge for Storage Networking and


Software Companies
Clustering could be a Clustering could be a huge opportunity for storage networking players. Clustering
huge opportunity for software has been one of Veritas’s fastest growing products with a 140% CAGR
storage networking
players.
(albeit off a small starting base) as illustrated in Figure 70 below.

74
The SAN Book III – October 22, 2001

Figure 70. Veritas Clustering Revenues


Revenue in millions

$70

$60
140% CAGR
$50

$40

$30

$20

$10

$0
1Q99 2Q99 3Q99 4Q99 1Q00 2Q00 3Q00 4Q00 1Q01 2Q01

Source: Company reports and Salomon Smith Barney

Clustering uses the Virtual Interface (VI) and, since InfiniBand is not available
today, Fibre Channel has become the preferred networking medium to run VI due to
Fibre Channel’s efficient design. That said, IP is the most proliferated protocol for
server clustering. This has created a natural partnership between VI and Fibre
Channel which, in turn, has opened up many storage networking companies to the
development of Infiniband in order to expand their addressable market. Intel has
invested in Ancor (owned by Qlogic), Crossroads, and other emerging networking
companies to promote the development of Infiniband. Startups such as Mellanox
and Banderacom are focused on developing next-generation Infiniband technology
and have also drawn the attention of Intel.
There is currently a ➤ Ethernet is not typically used to run VI as a result of IP’s inefficient protocol
strong partnership stack. We believe this is one of the largest drivers behind many Ethernet
between VI and Fibre
Channel, which has
networking companies’ efforts to upgrade the IP stack: to be able to cluster more
resulted in the use of efficiently.
Fibre Channel to network
large server clusters. ➤ Novell, Veritas (Cluster Server, Global Cluster Manager), and Legato
(Automated Availability Manager) have each released 32-node cluster software
packages with the ability to expand. While many sans may start small (under
ten server nodes), customers are taking advantage of existing scalable clusters
which can scale from two to an almost infinite number of systems. In the
second quarter of 2001, Veritas’s Cluster Server represented over 20% of its
licensed revenues.

75
The SAN Book III – October 22, 2001

This page intentionally left blank.

76
The SAN Book III – October 22, 2001

Subsystems

Subsystems: The Heart of Storage


Disk Drives to Subsystems
RAID and Beyond
NAS Subsystems
Core Software

77
The SAN Book III – October 22, 2001

z
Subsystems: The Heart of Storage

➤ While all the buzz is about newer growth areas of storage networking
and software, the reality is that subsystems are the heart of storage.

➤ In our view, as an investor, the ability to understand the fine nuances


of new subsystem technology developments can give you an edge in
helping to predict a company’s future success.

Subsystems Drive Demand


Several major subsystems vendors (such as EMC, Hitachi Data Systems, IBM,
Compaq, and Network Appliance) and startups (such as 3PARdata, 3ware, and
Broadband Storage) have recently released exciting new products that provide major
architectural upgrades by leveraging new component technology developments and
more advanced software functionality. For example, EMC has its HighRoad
architectures; Network Appliance has its DAFS initiatives; Hitachi has its Hi-Star
Switched Architecture; TrueSAN has its Meta-Fabric; 3PARdata has carrier-class
storage; 3ware has its ATA-based RAID controllers; and Blue Arc has its
SiliconServer architecture that utilizes field programmable array units (FPGAs).

The Heart of Storage


Storage subsystem Subsystems are the heart of storage. Storage subsystem purchases are an important
purchases are an catalyst for the migration to SANs since they provide much of the unique
important catalyst for the
migration to SANs since
functionality of this architecture and sales are directly related to capacity. SAN
they provide much of the RAID-based disk storage (including SAN-attached external storage and NAS)
unique functionality of represented 65% of the total SAN market in 2000 (see Figure 71 below).
this architecture and
sales are directly related Figure 71. SAN Storage Market in 2000
to capacity.

Storage
Software
22.2%

SAN
Networking RAID Storage
13.3% 64.5%

Source: Dataquest (August, 2001)

As subsystems and As subsystems and architectures improve in functionality, performance, and price,
architectures improve in we believe more end users will invest in SANs. As SAN penetration improves, the
functionality,
performance, and price,
adoption of complementary technologies such as storage networking and software
we believe more end should offer exceptional growth.
users will invest in
SANs. Figure 72 below illustrates the high growth that storage subsystems are expected to
experience.

78
The SAN Book III – October 22, 2001

Figure 72. RAID-Based Disk Storage Revenues and Projections


$ in millions

CAGR
1998 1999 2000 2001 2002 2003 2004 2005 2000-2005
Internal $4,762 $5,818 $7,064 $7,269 $7,223 $6,987 $6,427 $5,704 (4.2%)
JBOD $5,083 $5,704 $4,582 $3,887 $3,162 $2,459 $1,824 $1,289 (22.4%)
External (Subsystems) $12,093 $13,491 $16,330 $16,735 $19,270 $23,963 $31,327 $41,063 20.3%
NAS $349 $580 $1,448 $1,950 $3,521 $5,347 $7,502 $10,343 48.2%
Total RAID Revenue $22,287 $25,593 $29,424 $29,841 $33,176 $38,756 $47,080 $58,399 14.7%
Year-to-Year Growth -- 14.8% 15.0% 1.4% 11.2% 16.8% 21.5% 24.0%

Source: Dataquest (August, 2001)

Throughout the report, we use the following terminology:


➤ Internal: Both the storage and the RAID (Redundant Array of Independent
Disks) intelligence are inside the server.
➤ JBOD: Acronym for “Just a Bunch Of Disks” and that’s all it is. Although the
storage is externalized from the server, the RAID intelligence remains inside the
server.
➤ External, or Subsystems: Both the storage and the RAID intelligence are
outside the server. This is EMC’s market and the main focus of this report.
➤ NAS: Acronym for “Network-Attached Storage.” NAS is a dedicated storage
appliance optimized for file serving and storing data that connects to the existing
infrastructure (i.e., the LAN).
Figure 73 below shows the market shares of the total RAID-based storage market
including internal, JBOD, and external storage.
Figure 73. Total RAID Storage Market Share by Revenue

1998 2000

Hitachi, HDS, Other EMC


Other EMC Hewlett-Packard 14.9% 19.1%
19.4% 14.4% 14.0%

Compaq
20.0%

Hitachi, HDS, Fujitsu, Fujitsu Compaq


Hewlett-Packard Siemens 18.7%
13.9% 3.7%
Fujitsu, Fujitsu Sun IBM
Sun IBM Dell
Siemens 11.6% 13.3%
9.7% 15.2% 4.6%
4.1%
Dell
3.3%

Source: Dataquest (August, 2001)

Note: We have aggregated Hitachi and Hitachi Data Systems (HDS) with Hewlett-
Packard since the company OEMs HDS’s Lightning subsystem.

79
The SAN Book III – October 22, 2001

Since storage capacity is We believe the ability to understand the fine nuances of new subsystem technology
at the heart of the developments can provide an edge in helping predict a company’s future success.
industry, the subsystems
companies have
Also, since storage capacity is at the heart of the industry, the subsystems companies
tremendous influence on have tremendous influence on the direction of related technologies.
the direction of related
technologies. For example, in June, 1995, EMC introduced Symmetrix 3000, one of the world’s
first platform-independent storage subsystems, which included EMC’s first entrance
onto open systems platforms (i.e., NT and UNIX) in addition to its mainframe
platform. Although the significance of this technological development may seem
obvious in hindsight, it took about a year before the new product translated into
significant revenues since it had to target a new open system customer base. This
new product cycle fueled growth for EMC for over five years and forever changed
the dynamics in the storage industry (see Figure 74).
Figure 74. EMC’s Stock Price Versus the S&P

6000%
EMC
S&P

5000%
Platform Sales
independent product ramp up
release

4000%
Time to ramp selling to
Relative Performance

new open systems


customer base

3000%

2000%

1000%

0%
Ju 4

Ju 5

Ju 6

Ju 7

Ju 8

Ju 9
Se 3

D 3

Se 4
D 94

Se 5

D 5

Se 6

D 6

Se 7

D 7

Se 8

D 8

Se 9

D 9
M 3

M 4

M 5

M 6

M 7

M 8

9
-9

-9

-9

-9

-9

-9
-9

9
-9

-9

-9

-9

-9

-9

-9
p-

n-

p-

n-

p-

n-

p-

n-

p-

n-

p-

n-

p-
ar

ar

ar

ar

ar

ar
n

ec

ec

ec

ec

ec

ec

ec
Ju

Source: FactSet and Salomon Smith Barney

Understanding subsystem technology can help answer questions such as:


➤ What is it about EMC’s Symmetrix technology that has allowed it to maintain
its leadership position for over a decade? Why is it positioned so well for the
future?
➤ Why is Hitachi’s Hi-Star architecture such a success?
➤ What are the major product differences to focus on?
➤ In addition to hardware, how significant is the software to the performance of a
subsystem? (Our Answer: A lot!)
➤ What kind of innovations are startups developing?

80
The SAN Book III – October 22, 2001

➤ Why is a sales force and customer support (“the business”) so important?


With so many new subsystems products and terminology coming out recently, we
felt it would be useful to gain a better understanding of what they are and what
benefits they deliver. We start off with understanding the basics: What is a disk
drive? Then we move into the major trends in storage. Finally we discuss the
technology: How do different RAID algorithms affect performance? How does a
subsystem work? Why is cache so important? What is the difference between bus
and switched architectures? Read on to find out everything you need to know about
subsystems to be dangerous.

81
The SAN Book III – October 22, 2001

Disk Drives to Subsystems


➤ Increasing demand for external storage subsystems has opened up
the storage market and enabled externally focused companies such
as EMC to compete in more of the market.

What Is a Disk Drive?


Did you know disk drives Did you know disk drives can account for as much as 60% of the cost of a subsystem?
can account for as much On a cost basis, the disk drive can be considered the most important component in a
as 60% of the cost of a
subsystem?
subsystem. Also, disk drives can affect the speed and overall performance of a storage
subsystem. Hence, in understanding subsystems, it is important to understand what
disk drives are and how they work. The major disk drive companies are Maxtor,
Seagate, IBM, Western Digital, and Fujitsu.
When you want to store information on your computer so you can retrieve it later, you
can either save it to a removable disk (i.e., floppy, CD, or zip) or an internal hard disk
drive (HDD, or simply, “disk drive”). The difference between a disk drive and a
floppy disk is that a disk drive is fixed inside the computer, it’s faster, holds more
information, is more reliable, is more expensive, is larger, and has higher areal density
(how much data can be stored per square inch). Nearly all PCs have a single disk
drive built inside the computer, whereas disk storage subsystems (“subsystems”) have
multiple, sometimes hundreds, of disk drives. There are three types of disk drives: 1)
IDE, 2) SCSI, and 3) Fibre Channel. In this report, we address enterprise drives,
which include SCSI and Fibre Channel, since those are used for disk storage
subsystems.
Figure 75 below illustrates the inside of a disk drive.
Figure 75. Major Hard Disk Drive Components

Source: Seagate Technology; Note “Disc” is also referred to as “Disk” or “Platter”.

82
The SAN Book III – October 22, 2001

While disk drives vary, they all contain the same basic components. In many ways,
disk drive components are analogous to those in an audio record player: the disk
platter is analogous to the record, the head is analogous to the needle, and the
suspension/head arm is analogous to the record player arm.
Like an audio record, information is recorded on both sides of the platter. However,
in a disk drive the disk platter is fixed in place. Thus, there is a head (i.e., a needle)
on each side of the disk platter (i.e., record). Also, in a disk drive there is typically
more than one disk platter per drive. Since each disk platter can contain a fixed
amount of data, more disk platters can be stacked into a disk drive to expand its
storage capacity. Disk drives can have up to 12 platters and can be re-recorded over
(much like a tape cassette).
The disk drive in Figure 75 above has three media and six heads. Assuming it stores
33.3 GB on each platter, it will hold 100 GB of information.
Disk drives typically In Figure 76 below, to the left is what a single disk drive looks like as an internal
account for about 40% to component and to the right what it looks like as a stand-alone external product. Disk
60% of a storage
subsystem’s cost.
drives typically account for about 40% to 60% of a storage subsystem’s cost. Note:
While the disk drive on the left displays its internal architecture, disk drives in use
are fully enclosed to look like a small rectangular pizza box.
Figure 76. Disk Drives

Internal External

Source: Maxtor

Figure 77 below has an example of where a disk drive could be located within a
subsystem. In this case, we use EMC’s Symmetrix as an example of a subsystem.

83
The SAN Book III – October 22, 2001

Figure 77. EMC Symmetrix Disk Drive Subsystem

Fans
(outtake)

RAID
(Intelligence)
Disk
Drives

Control
Console

Power
Logic and
Memory Boards
Fans
(intake)

Source: EMC and Salomon Smith Barney

Moving from Internal → External Storage


There are three basic configurations for RAID-based storage systems: 1) internal, 2)
JBOD, and 3) external.
Figure 78 below illustrates these configurations. In the internal storage
configuration, the server, RAID intelligence, and drives are housed within the same
chassis. While it has a simpler design with tighter integration, internal storage also
has its limits including scalability, the ability to mix best-of-breed server and
storage, fault tolerance, and reliability. In the JBOD configuration, the server and
RAID intelligence are tightly integrated but the disk drive is externalized. While
JBOD is a step in the right direction, it separates the intelligence from the data.
Subsystems integrate Subsystems solve these problems by integrating the RAID intelligence and the disk
the RAID intelligence and drives in a single chassis and externalizing the subsystem from the server.
the disk drives in a
single chassis and
Additionally, subsystems such as EMC’s Symmetrix have multiple ports so data
externalize the from several servers can be aggregated into a single storage pool and as a result,
subsystem from the become easier to manage. In this section we are primarily concerned with internal
server. versus external storage. More details about JBOD appear under “RAID and
Beyond”; and more on networked external storage is included in the “What is a
SAN?” section.

84
The SAN Book III – October 22, 2001

Figure 78. Different Server/RAID Configurations

JBOD Configuration

Internal Server RAID Drive

Server RAID Drive


JBOD Networked External
External Storage Configuration Storage Configuration

Server RAID Drive Server RAID Drive

Network Subsystem
Subsystem

Source: Salomon Smith Barney

Server vendors have Server vendors are typically more motivated to promote internal versus external
several advantages when storage since they have several advantages when it comes to selling storage inside
it comes to selling
storage inside the server
the server including brand name familiarity and initial point of contact with the
including brand name customer.
familiarity and initial
point of contact with the Figure 79 below depicts internal versus external storage. The thick line connecting
customer. server and storage in the external configuration can represent a SCSI or Fibre
Channel cable. More complex configurations can be created with switches for
networked storage.
Figure 79. Internal Versus External Storage

Internal External

Storage

Server Server Storage


Source: Salomon Smith Barney

The obvious result is that many of the top internal storage vendors are also server
vendors, as illustrated in Figure 80 below.

85
The SAN Book III – October 22, 2001

Figure 80. Internal RAID Storage Market Share by Revenue

1998 2000
Fujitsu, Fujitsu
Siemens
Fujitsu, Fujitsu 2.1%
Siemens Other Other
Compaq NEC
4.6% 15.4% 17.1% Compaq
29.3% 1.4%
30.3%
NEC
Hewlett Packard
2.0%
7.1%
Hewlett Packard
7.5% Sun
Sun 7.6%
Dell IBM Dell IBM
8.1%
7.6% 25.5% 12.3% 22.2%

Source: Dataquest (August, 2001)

When players such as The server vendors were able to retain a large percentage of storage market share for
EMC and StorageTek years due to its internal design. However, when players such as EMC and
entered the storage
market and
StorageTek entered the market and demonstrated the benefits of external storage, a
demonstrated the new segment emerged. Some of the benefits of external storage include better
benefits of external scalability, functionality, reliability, flexibility, serviceability, and less CPU cycles.
storage, a new segment Unlike the server vendors, storage vendors, such as EMC, were more motivated to
emerged.
sell external storage since this was the only type of storage they sold, rather than
keeping it captive to their server sale by bringing it inside the server (or “internal”).
Many server vendors evolved by offering JBOD. But while this appeared to be
“external” storage, it was not. Server vendors succeeded to some degree with this
initiative until the advent of storage area networking (SAN), which highlighted the
benefits of external storage. As a result of its numerous advantages, external storage
has been growing quickly while internal storage and JBOD revenues have begun to
decline.
From 1997 to 2000, EMC’s organic revenues grew at a compound annual growth
rate (CAGR) of 37.9%. Since it is the clear leader and only participates in the
external storage subsystems market, it is a great real-world illustration of the growth
of its market.
Figure 81 below demonstrates EMC’s Symmetrix revenue stream — with a CAGR
of 26.3% from 1995 to 2000 — as a great example of the success of external
subsystems.

86
The SAN Book III – October 22, 2001

Figure 81. EMC’s Symmetrix Revenues

Revenue in millions
$1,600

$1,400
CAGR 26.3%
$1,200

$1,000

$800

$600

$400

$200

$0
1Q95

3Q95

1Q96

3Q96

1Q97

3Q97

1Q98

3Q98

1Q99

3Q99

1Q00

3Q00

1Q01
Source: Company reports and Salomon Smith Barney. CAGR is based on annual revenues from 1995 to 2000 and does not include 2001.

Figure 82 below demonstrates that although the major server vendors own the
internal storage market, EMC and Hitachi Data Systems (pure play storage vendors)
are the market leaders in the external storage market. Note: We have aggregated
Hewlett Packard with Hitachi Data Systems and Hitachi since HP OEMs Hitachi’s
product.
Figure 82. Internal Storage Versus External Storage Market Share by Revenues (in 2000)

Internal Storage External Storage

Fujitsu, Fujitsu Dell PowerVault


Siemens 1.0%
2.1% Other
NEC
Other 14.7%
NEC 2.0% EMC
17.1% Compaq
1.4% 32.7%
30.3% Fujitsu
Hewlett Packard 5.6%
7.1%

Sun
7.6% Sun Network Storage Compaq Storage
Dell IBM 7.1% IBM 11.3%
Hewlett Packard,
12.3% 22.2% 8.4%
Hitachi and HDS
17.2%

Source: Dataquest (August, 2001)

However, server vendors still have an advantage to selling and a desire to sell
JBOD. Figure 83 below illustrates that for JBOD storage, the storage is captive (the
brand of the JBOD is the same as the brand of the server) 99% of the time.
However, external storage subsystems are captive in less than half of the cases and
declining.

87
The SAN Book III – October 22, 2001

Figure 83. JBOD Storage Versus External Storage by Revenues in 2000

JBOD External Storage

Noncaptive
0.6%
Captive
44.4%

Noncaptive
55.6%

Captive
99.4%

Source: Dataquest (August, 2001)

Figure 84 below demonstrates that increasing demand for external storage


subsystems has opened up the storage market and enabled externally-focused
companies such as EMC to compete in more of the market.
Figure 84. Total RAID-Based Disk Storage
Revenue in millions

$45,000

$40,000
$35,000

$30,000
External
$25,000
CAGR 20.3%
$20,000

$15,000
Internal
$10,000
CAGR (4.2%)
$5,000
JBOD, CAGR (22.4%)
$0
1998 1999 2000 2001 2002 2003 2004 2005

Source: Dataquest (August, 2001). CAGR is for 2000–05.

Since server vendors have less of an advantage selling external storage than they do
internal storage, market share position can change quickly. Figure 85 shows the
market share for external storage for 1998 and 2000. Market share percentages
change more in external storage than in internal storage (see Figure 85 below). The
more dynamic nature of the external market gives independent vendors such as LSI
Logic and startups a better opportunity to penetrate the market.

88
The SAN Book III – October 22, 2001

Figure 85. External RAID Storage Market Share by Revenues (1998 and 2000)

1998 Dell PowerVault 2000


1.0%
NEC Other
Other EMC
2.0% 14.7%
23% 26% EMC
Fujitsu 32.7%
NEC 5.6%
2%

Fujitsu
Compaq Storage
7%
8%
Hitachi, HDS Sun Network Storage Compaq Storage
IBM IBM
13% 7.1% Hewlett Packard, 11.3%
11% 8.4%
Hitachi and HDS
Sun Network Storage
Hewlett Packard 17.2%
6%
4%

Source: Dataquest (August, 2001)

89
The SAN Book III – October 22, 2001

RAID and Beyond


➤ RAID is an algorithm (formula) that describes a method of writing
data to a group of disks to improve performance and protect against
disk failure. Which RAID algorithm a subsystem vendor decides to
offer can impact the design, cost, reliability, and speed of its
subsystem.

RAID-ing Scale of 1 to 5
Which RAID algorithm a In 1987, researchers David Patterson, Garth Gibson, and Randy Katz published a
subsystem vendor paper that outlined the concept of RAID (Redundant Array of Independent Disks).
decides to offer can
impact the design, cost,
RAID is an algorithm (formula) that describes a method of writing data to a group of
reliability, and speed of disks to improve performance and protect against disk failure. Which RAID
its subsystem. algorithm a subsystem vendor decides to offer can impact the design, cost,
reliability, and speed of its subsystem. For example, a new subsystem designed
using RAID 5 would likely have greater capacity but slightly less speed than a
subsystem using RAID 1. EMC’s RAID expertise has helped propel it to the leading
position within the market. In this section we describe what these terms mean, the
different RAID levels available, and their relative advantages and disadvantages.
What Is RAID?
A subsystem with RAID intelligence inside is smarter, more reliable, and more
expensive than JBOD (Just a Bunch Of Disks) which separates the intelligence (i.e.,
RAID) from the disk. A RAID subsystem uses algorithms to write data to a group
of disks so that the drives’ combined performance and reliability are better than if
they were working independently. Several RAID algorithms exist, each with its
own advantages. The methods differ primarily in the way they write/retrieve data
and build in fault tolerance. We will explain two of the more popular methods in
detail, RAID 1 and RAID 5, and then touch upon a few others.
RAID 1: Mirroring
Mirroring is more fault- In the RAID 1 algorithm, data is written to disk using mirroring. In mirroring, every
tolerant, but takes up a time data is written to disk, an exact copy is made to a different disk in the same
lot of disk space.
subsystem (Figure 86 below). This is accomplished by writing the data to two
different drives at the same time.

90
The SAN Book III – October 22, 2001

Figure 86. Mirroring in RAID 1

123 123

Source: Salomon Smith Barney

If one drive fails, an exact copy exists on a separate healthy drive and no data is lost.
When the original failed drive is replaced with a new drive, the mirrored drive (the
backup copy) becomes the primary copy and the new drive becomes the backup
(Figure 87 below). Therefore, the subsystem always has two copies of the data,
thereby ensuing that data is not lost.
Figure 87. Redundancy in RAID 1

X 123

123
Co
py
123

Replacement Disk

Source: Salomon Smith Barney

In addition to improving fault tolerance, RAID 1 also enables faster read times. If
two computers are trying to access the same file simultaneously, rather than waiting
in queue, they can each access a different copy of the data. But, there are no free
lunches. The main disadvantage to RAID 1 is that it is expensive since it requires
twice the storage capacity of a non-mirrored configuration.

91
The SAN Book III – October 22, 2001

RAID 5: Striping with Parity


Striping divides a piece The RAID 5 algorithm uses striping — a technique in which data received is
of data and each piece is segmented and each segment is written to a different drive — to write to disk. For
written to a different
drive. It makes writing
example, the first block would be written to the first drive, the second block to the
and reading to disk second drive, and so on (Figure 88 below). It makes writing and reading to disk
faster. faster.
Figure 88. Striping

No Striping Striping
123 123

123 1 2 3

Source: Salomon Smith Barney

RAID 5 builds in fault RAID 5 builds in fault tolerance using fewer disks than RAID 1. The equivalent of
tolerance using fewer one disk of storage is used to store parity per each RAID set of disks.
disks than RAID 1. The
equivalent of one disk of Figure 89 below demonstrates how parity works. The RAID controller or software
storage is used to store calculates the data in the first row (the numbers 1, 2, and 3) add up to 6, which is
parity per each RAID set
of disks. then stored on a different disk. Note: In actual practice RAID 5 uses 1s and 0s for
the data, the parity calculation is stored across several disks instead of one, and the
parity calculates if the data on the other disks add up to an odd or even number.
Figure 89. Parity in RAID 5

1 + 2 + 3 = 6

Data Parity

Source: Salomon Smith Barney

In RAID 5, if one of the disks fails, information from the other disks can be used to
recreate the data on a spare disk. In Figure 90 below, according to the parity disk,
the data on the other disks should add up to six. By subtracting one and three from

92
The SAN Book III – October 22, 2001

six, the RAID 5 intelligence can determine a two was on the failed disk and thus
recreate it.
Figure 90. Redundancy in RAID 5

X + 0 + 3 = 6

(=6-1-3)

Replacement Disk

Source: Salomon Smith Barney

The benefits of RAID 5 are that it can decrease the time it takes to write since it
spreads the workload across drives and it is less expensive since it takes up less disk
space than RAID 1, but again, there are no free lunches. It also requires greater
CPU utilization in the subsystem, thereby requiring more silicon due to the greater
level of difficulty in calculating the more complex RAID 5 algorithm versus a RAID
1 algorithm. It also takes longer to rebuild due to the calculations involved.
RAID 5, however, Is RAID 5 as reliable as RAID 1? No. In a situation in which multiple drives fail at
requires greater CPU the same time, RAID 5 can actually lose data. For example, in the rare event that
utilization in the
subsystem and takes
drives 1 and 2 went down simultaneously before either could be replaced, parity
longer to rebuild. cannot be used to calculate what data was stored on each drive.
Other RAID Levels
RAID levels 2, 3, and 4 also stripe and use parity schemes slightly different than in
RAID 5. RAID 0+1 and RAID 10 use both striping and mirroring. These RAID
levels are not used as often as RAID 1 or 5. Also, EMC uses its proprietary RAID S
for its Symmetrix flagship subsystem. In RAID S, part of the RAID calculation is
performed by the disk drives themselves instead of by the disk controller. RAID S
does not use striping, but it is often compared with RAID 5. Many vendors such as
EMC provide customers with choices of multiple RAID levels.
Figure 91 below charts what RAID levels certain vendors incorporate in their
subsystems.

93
The SAN Book III – October 22, 2001

Figure 91. RAID Vendor Comparison

1 2 3 4 5 S 0+1 10
Compaq StorageWorks 12K FC • • • •
Dell PowerVault 660F • • • •
EMC Clariion • • • •
EMC Symmetrix • • •
HDS Lightning 9900 • •
HDS Thunder 9200 • • •
IBM ESS (Shark) •
IBM MSS • • • •
LSI Logic e4400 • • • •
Network Appliance •
Sun StorEdge T3 • •
Source: Salomon Smith Barney

Several interesting patterns can be observed in the figure above:


➤ RAID 1 is the most often used in the enterprise market (i.e., high end), and
RAID 0+1 and RAID 10 are also popular choices. EMC has built its success on
RAID 1 due to its reliability. RAID 5 is the second most commonly used and
tends to show up more in the mid-range or low end of the market.
➤ Network Appliance is one of the few vendors that uses RAID 4, which is similar
to RAID 5 but stores all parity calculations on a single dedicated disk instead of
several disks. Network Appliance chose to use RAID 4 in order to ease the
process of adding disks to an existing RAID group. Although this can slow
down writing and reading, Network Appliance believes it has optimized around
this with its file system.
➤ RAID 3 is popular for video applications.

Data’s Path to Performance


The method of The method of transferring data from the network to the disks within a subsystem
transferring data from can have a significant impact upon performance. Many variables exist such as the
the network to the disks
within a subsystem can
size of the cache, the effectiveness of the caching algorithms, and whether a bus or
have a significant impact switched architecture is used. These can affect how quickly data is processed,
upon performance. stored, and retrieved by the subsystem. Hence we believe it is important for
investors to understand what path the data takes in a subsystem in order to
understand how a system might ultimately perform.
Figure 92 below illustrates the basic path data takes in a subsystem. We will use a
read request as an example, although a write command would take a similar path.
First, the host computer will make the request (either read or write) to a subsystem.
Along the path, it stops at several points within the subsystem for either processing
or transport (see Figure 92 below).

94
The SAN Book III – October 22, 2001

Figure 92. Subsystem Architecture

1 2&3 4 5 6

SCSI Backplane Ctr


Interf lr
Cache Bus
FC Ctr
or lr
Interf
Logic Ctr
IP
Fabric lr
Interf
Disk
Disks
controllers

Battery Power Supplies Fans


Connectivity
Server Storage Subsystem (External)

Source: Salomon Smith Barney

The data takes the following path, in order:


1 Interface: The request command will typically travel from the server, across a
SCSI cable or Fibre Channel network, and into the storage subsystem. It enters
the subsystem through an interface. The interface consists of a port and a target
controller adapter (or chip) which contains intelligence. The interface translates
the electrical signals that it receives from the external cable into a digital form
which the computer can understand. A different adapter is needed for each
protocol (i.e., SCSI and Fibre Channel). SCSI and Fibre Channel are the two
most commonly used protocols for communications between servers and storage.
2 Logic: Next, a microprocessor processes the logic. The logic is responsible for
several functions including checking if the data requested is readily available
from the cache and keeping track of which data has been requested
3 Cache: Next, the cache checks to see if it has the data requested. Cache uses
silicon instead of magnetic media to store data. Since it is made up of silicon
instead of spinning magnetic platters, it has a quick access time. If the cache
does not have the data stored, it will send the request along to the backplane.
(We describe the cache in greater detail in the “Cache Costs Ca$h” section.)
The backplane typically 4 Backplane: Next, the request for data will travel across a backplane. This part
uses a bus or switched
of the subsystem is used to interconnect one section to another to transport data.
architecture.
The backplane typically uses a bus or switched architecture. In figure 92 above,
it is used to transport data between the cache and the disk controllers. (We
describe the main differences in performance between these two architectures in
a section titled “Bus Versus Switched Architecture.”)

95
The SAN Book III – October 22, 2001

5 Disk Controller: Next, the data goes to the disk controller. The disk controller
is a set of chips that contains the intelligence to manage the actual writing and
reading to disk.
6 Disk Drive: Next, the data is stored in the disk drive. Once the data is found
and retrieved, it will be sent to the host that initially requested it in nearly the
reverse order.
Subsystems use Subsystems use different combinations and architectures of hardware and software
different combinations to execute similar functions. Figure 93 below provides a different view. It gives an
and architectures of
hardware and software
idea of how various subsystems, which can be largely composed of off-the-shelf
to execute similar components, can still be unique in terms of performance and design.
functions.
Figure 93. Subsystems Architecture Model

Data Path Variables

(1) Interface SCSI, FC

microprocessor,
(2) Logic
algorithm, firmware

(3) Cache size, mirroring

(4) Backplane bus or switch

(5) Disk Controller RAID level

(6)
Disk SCSI, FC, ATA

Source: Salomon Smith Barney

In architecting a subsystem, several options exist:


1 Interface: A vendor can decide which protocols it wants to support (e.g., SCSI,
Fibre Channel).
2 Logic: A vendor can choose which microprocessor it uses. For example, EMC’s
Symmetrix uses the 333 MHz PowerPC (up to 80) while Hitachi Data System
(HDS) uses Intel’s I960.
3 Cache: A vendor can decide on size and if it wants to mirror the cache (for more
on this see “Cache Costs Ca$h”).
4 Backplane: A vendor can decide between using a bus or a switched architecture.
Backplanes can also be either active or passive.
5 Disk controller: A vendor can decide which RAID level it wants to support.
There can be more than one RAID level supported in a single subsystem.

96
The SAN Book III – October 22, 2001

6 Disks: A vendor can choose among different interfaces (i.e., Fibre Channel,
SCSI), speeds, and capacities.
In any design, there are There are many considerations to choose from when designing a subsystem. In any
trade-offs to using one design, there are trade-offs to using one method versus another to fulfill a function.
method versus another
to fulfill a function.
One of the more common trade-offs is price versus performance. For example, a
vendor could use the best hardware available to design a subsystem, but it would
likely be expensive to manufacture. Additionally it still might not perform as well
as an older system which had more advanced firmware.
Another example of a subsystem consideration is “to make” versus “buy.” For
example, Hewlett-Packard OEMs its high-end subsystem from Hitachi Data
Systems, allowing it a faster time to market. On the other hand, IBM has chosen to
design its own subsystem, the “Shark” (although in the past it has OEMed from
StorageTek), giving it higher potential margins and greater control over future
product releases.
Although knowing the specifications of a new subsystem can help predict a
subsystem’s performance relative to competitors, the ultimate test is still customer
adoption and sales. Customers choose vendors on many factors in addition to
“speeds and feeds” such as support, software applications, price, reliability, ease-of-
use, functionality, scalability, size, interoperability, and total cost of ownership
(TCO). Since different customers have different needs, companies that can satisfy
the broadest set of needs in the most complete way should gain the most market
share.

Cache Costs Ca$h (but Looks Worth Every Penny)


Cache is a fast but Some vendors such as EMC claim their caching is one of their biggest advantages
expensive way to store over other vendors’ products, and we agree! Although it is typically Static Random
data. It is typically used
for fast retrieval of
Access Memory (SRAM) which is widely available, the way cache is implemented
frequently-accessed and utilized in the subsystem can dramatically affect performance. Caching can
data. speed up performance through more intelligent algorithms and increased memory
capacity. Fault tolerance can be improved through mirrored caching and battery
backup.

What Is Cache?
Cache (pronounced cash) stores data in silicon instead of magnetic drives. As a
result, it is faster and more expensive. A megabyte of storage capacity on a hard
disk drive can cost around $0.03, but on silicon such as SRAM, can cost $10. Think
of it as short-term memory (cache) versus long-term memory (disk drive). Data in
cache is easier and faster to retrieve but is not stored as long as data in disk drives.
Subsystems use cache to store the most frequently accessed data so the total amount
of time it takes to access data decreases. The amount of time it takes to access cache
is measured is nanoseconds (billionths of a second) whereas the amount of time it
takes to access disk drives is measured in milliseconds (thousandths of a second).
In addition to having a faster access time versus disk drives, getting data from cache
will also shorten the path the request must make in the subsystem before retrieving
the data. If the subsystem looks for the data and the cache already has it, the

97
The SAN Book III – October 22, 2001

subsystem can retrieve the data directly from cache without ever having to access
the disk drive. However, if it does not find the data in the cache, the subsystem must
then go to the disk drive.
Figure 94 below illustrates a cache hit (requested data is in the cache) versus a cache
miss (requested data is not in the cache).
Figure 94. Cache Hit Versus Cache Miss

Cache Hit: No disk access needed

1
CPU Cache Disk

2
Cache Miss: Need to access disk

1 2
CPU Cache Disk

4 3
Source: Salomon Smith Barney

Cache is analogous to closet space. In order to save time and energy, you would
probably store the outfits you wear most often in the closet nearest to you, the
bedroom closet (cache), rather than one further away, the attic closet (disk). Now
imagine that on a subsystem scale — you would only be able to store three shirts out
of 100 in that nearby closet (analogous to 32 gigabytes out of one terabyte).
There are several ways to improve the use of cache including 1) increasing the size
of the cache, 2) improving the algorithm used to decide which data is stored in
cache, 3) mirroring, and 4) battery backup. Increasing the size of the cache is
analogous to having a bigger bedroom closet. The more you can store in the
bedroom closet (cache), the less often you have to go to the attic (disk). Most
current subsystems have cache sizes ranging from several megabytes to, for
example, 64 GB on EMC’s Symmetrix. We will describe the other three variables
in greater detail next.
Speed Through Caching Algorithms
A caching algorithm is a This is one of EMC’s greatest competitive advantages, in our view. A caching
formula that tells the algorithm is a formula that tells the subsystem which data to store in cache. If the
subsystem which data to
store in cache.
cache is one gigabyte in size, the algorithm’s goal is to correctly guess which
gigabyte of data will be requested next. Returning to the closet analogy, the caching
algorithm is similar to the way you guess which clothes you will likely wear next
week so you can determine what to put in the bedroom closet. Well-constructed
algorithms can speed up the overall subsystem by maximizing the amount of data
that can be retrieved from the cache (bedroom closet) so there is less time needed to

98
The SAN Book III – October 22, 2001

access the slower disk drives (attic closet). Note: The selection of which data is
stored in the cache changes continuously.
There are two types of There are two types of algorithms: read and write. Read algorithms determine what
algorithms: read and to store in the cache. Write algorithms determine when to write the data from cache
write.
to disk. A simple example of a read algorithm is “store the data that you used most
recently in the cache.” Good caching algorithms are incredibly difficult to create
since they are essentially trying to predict the future.
Figures 95 below shows how the cache relates to accessing files.
Figure 95. Algorithms Guess Which Files Will Be Accessed Soon

•The greater the overlap,


Files that the faster the data is retrieved.
will be Cache •A good algorithm
accessed soon means more overlap.

Overlap includes files that


do not require disk access

Source: Salomon Smith Barney

Figure 96 below graphically depicts an effective versus an ineffective caching


algorithm.
Figure 96. Effective Versus Ineffective Caching

Effective Ineffective

Source: Salomon Smith Barney

EMC’s Intelligent Caching Algorithms


We believe EMC has one of the best caching algorithms available as a result of its
decade-long focus in the area. In some instances, EMC’s Symmetrix storage

99
The SAN Book III – October 22, 2001

subsystem is able to fulfill approximately 90%–95% of the read requests it receives


from the cache without ever having to access a disk drive. This, of course, results in
EMC’s renowned performance. The Symmetrix has the ability to customize its
algorithms based on past user behavior (i.e., it learns its user’s style) while many
other subsystems simply use the same algorithm every time, regardless of the user’s
style.
Safety Through Mirroring and Battery Backup
Once the subsystem receives data from the host, it travels along the data path and
traverses through the cache before it is written to disk. Along the way, there can be
problems sending the data from the cache to the disk for a number of reasons
including cache failure and power loss. In the case of cache failure, the cache board
stops functioning, thereby losing the data stored on it before it was written to disk
(i.e., data is lost forever). Any data loss is considered unacceptable in storage
environments.
Cache data loss can be Cache data loss can be prevented through the use of mirrored caching. Figure 97
prevented through the below compares single versus mirrored caching. In a single caching scheme, the
use of mirrored caching.
data is transferred to a single cache board and then to disk. In a mirrored caching
scheme, the data is transferred to two cache boards (similar to RAID 1) so if one
fails, the other cache board has a copy and data is not lost.
Figure 97. Mirrored Cache Versus Single Cache in Cache Failure

Single Cache Mirrored Cache

Data Data

X
Cache
X
Cache
Mirror
Cache

Subsystem Subsystem

Source: Salomon Smith Barney

A second risk to data loss on the cache is power. The type of memory used for
cache (often SRAM) is usually volatile. In volatile memory, when there is a loss of
power, the data is lost forever. This can happen if there is a power blackout, the
subsystem power supply system fails or maybe somebody walks by a subsystem and
accidentally kicks the power cord out of the wall (which happens). In this scenario,
a battery can be used to supply power temporarily. This can give the subsystem
enough power to finish the process of transferring the data from cache to disk safely.
Even backup batteries are often duplicated, just to be safe (see how crazy storage

100
The SAN Book III – October 22, 2001

people can be). As a magnetic medium, disk drives do not need continuous power to
retain data so once the data is transferred to disk, it is safe even if the battery
completely depletes or fails.

Bus Versus Switched Architecture


Hitachi’s Hi-Star switched architecture has been recently gaining traction in the
subsystem industry. By using a switched versus bus backplane, Hitachi claims its
Lightning 9900 product can have 6.4 GBps internal bandwidth, higher than most
subsystems available on the market today.
First, What Is a Backplane?
A backplane is a printed circuit board (PCB) that interconnects the separate
components within a subsystem. One example of a backplane is the motherboard in
your PC onto which various components (i.e., microprocessor, memory, NIC, HBA,
etc) are mounted and connected.
A bus is a system of interconnects (mostly metal traces) run across the backplane.
Components send data to each other by using electronic signals along these metal
traces (wires can also be used).
Many subsystems Many subsystems currently use a bus architecture on the backplane in which several
currently use a bus devices share a common bus. In Figure 98 below, a bus is used to allow the
architecture on the
backplane in which
microprocessors to communicate with the cache boards, and the cache boards to
several devices share a communicate with the disk controllers. Note: A bus and a bus architecture describe
common bus. two different concepts, although a bus architecture utilizes a bus.
Figure 98. Bus Architecture

Bus Ctr
SCSI Bus
lr
Interf
Ctr
Logic Cache lr

FC Ctr
Interf lr

Battery Power Supplies Fans


Source: Salomon Smith Barney

A subsystem with a bus architecture can increase its bandwidth by utilizing more
buses.

101
The SAN Book III – October 22, 2001

Hitachi’s Hi-Star Switched Architecture


In a switched Hitachi Data Systems has recently replaced its bus-based storage systems with its
architecture, each Hi-Star switched architecture, giving it 6.4 GBps of internal bandwidth, one of the
component has a direct
connection to every
highest in the industry. The higher bandwidth is primarily a result of the point-to-
other component. point direct connection. In a switched architecture, each component has a direct
connection to every other component.
In a bus architecture, components share one common interconnect. If two sets of
components are trying to communicate with each other, they must share the
bandwidth available to the single interconnect. For example, if the interconnect is
capable of 1 GBps, when two components are sending data simultaneously, each one
is allocated half of the available bandwidth, or 500 MBps. In a switched
architecture, each component can send data at 1 GBps simultaneously.
Figure 99 below illustrates these two architectural differences. In the bus
architecture, Cache A and Cache B must share the same interconnect to get to either
disk controller. In the switched architecture, there are multiple paths between each
cache and each disk controller.
Figure 99. Switched Versus Bus

Bus Switch

Cache A Cache B Cache A Cache B

Disk Ctrl A Disk Ctrl B Disk Ctrl A Disk Ctrl B

Source: Salomon Smith Barney

The main advantage of a switched architecture is that it generally has higher


bandwidth than a bus architecture; however, again, there is no free lunch. The
components and the complex manufacturing process used to create a switched
backplane make it more expensive.
Figure 100 below illustrates where the Hi-Star switched architecture is used inside
Hitachi’s Lightning 9900 subsystems. It is used for communication between the
microprocessors, cache boards, and the disk controllers. Note: Each interface chip
has its own dedicated logic (i.e., processor) so there is no bottleneck as depicted
below.

102
The SAN Book III – October 22, 2001

Figure 100. Switched Architecture and Mirrored Cache

Ctr
SCSI
lr
Interf
Switch Cache Switch
Ctr
Logic lr

FC Cache Ctr
Interf lr

Battery Power Supplies Fans


Source: Salomon Smith Barney

103
The SAN Book III – October 22, 2001

NAS Subsystems
➤ NAS subsystems differ from traditional subsystems due to their
connectivity to the existing infrastructure (LAN) rather than an
emerging dedicated storage infrastructure (SAN).

What Is a NAS Subsystem?


NAS subsystems differ from traditional subsystems due to their connectivity to the
existing infrastructure (LAN) rather than an emerging dedicated storage
infrastructure (SAN). In this section we segment NAS into three categories: 1) high
end, 2) midrange, and 3) low end.

NAS Configurations
In Figure 101 below are a couple of NAS configurations. In a traditional NAS
appliance, the server, RAID intelligence and drives are all within the same box. In
EMC’s Celerra, a general-purpose server is replaced with a file server (thin server
optimized for file serving), which is then attached to EMC’s Symmetrix subsystems.
Figure 101. Several NAS Configurations Are Possible

NAS EMC Celerra

File File
RAID Drive RAID Drive
Server Server
Source: Salomon Smith Barney

High-end NAS
This is the original NAS market, pioneered by Auspex and now dominated by
Network Appliance. According to Dataquest, the high-end segment constituted the
largest revenue segment in 2000.
According to Dataquest, According to Dataquest, high-end or enterprise NAS devices are server appliances
high-end NAS devices for high-performance, high-capacity file sharing either as a resource for use by
are server appliances for
high-performance, high-
networked clients or within a SAN. They typically have RAID support, often have
capacity file sharing multiple storage controllers, and support multiple networking topologies such as
either as a resource for 100Base-T, Fibre Channel, FDDI, and ATM. Pricing is typically over $100,000
use by networked clients with capacities scaling to 6 TB.
or within a SAN.
Other definitions suggest that high-end NAS appliances include connectivity to
several network file systems, sometimes extending to proprietary systems and
mainframes, and performance is improved through automatic load and path
balancing and high-speed transfers between storage domains assigned to different
file or application servers. Accessibility is enhanced by a feature that ensures
disaster protection referred to as High Availability (HA). Failing components are
diagnosed, reported to a remote service site, and eventually corrected or replaced

104
The SAN Book III – October 22, 2001

before they become inoperable. Storage management features extend beyond


backup, to include applications such as archiving, library management, and
hierarchical storage management (HSM). Examples of high-end NAS devices
include Network Appliance’s Filer F800 series and EMC’s Celerra. EMC’s
Celerra/Symmetrix combination is an enterprise NAS solution (although we
consider it a SAN/NAS hybrid).
Network Appliance’s Filer
Although Network Appliance may not have been the sole inventor of the NAS
concept, it has certainly been considered the marquee NAS company and has made
its filers to be synonymous with NAS. Network Appliance’s strategy to remove the
I/O burden of file accessing from application servers with a separate dedicated file
storage to an appliance that connects directly onto the Local Area Network (LAN)
has developed a storage market that has caught the eyes of investors and competing
storage vendors alike.
We believe Yahoo uses Network Appliance’s Filers have quietly been increasing their features/functions and
over 500 TB of Network
migrated into the data centers of many corporations and ISPs such as Cisco, Yahoo,
Appliance storage —
only eight IT Texas Instruments, Motorola, British Telecom, and Deutsche Telekom. We believe
administrators are Yahoo uses over 500 TB of Network Appliance storage for its Web-based email and
needed to manage it. account information and needs only eight IT administrators to manage it.
Figure 102 below shows Network Appliance’s F840. It appears in the foreground as
a single unit while the background shows it as part of a rack configuration.
Figure 102. Network Appliance’s F840

Source: Network Appliance

EMC’s Celerra — High-end NAS


The EMC Celerra File Server is a network-attached storage system providing high
performance, high availability and scalability for enterprise file storage. The system
features EMC-developed software and hardware, known as Data Movers, within a
common enclosure and with a common management environment. The Data

105
The SAN Book III – October 22, 2001

Movers and management environment connect to Symmetrix Enterprise Storage


systems and external customer networks through a variety of interfaces. A file
server’s purpose is to provide end users with the ability to access, update, and store
common files to a central location directly from their desktop computer without the
need to utilize a general purpose server. The company intends to continue to
introduce new versions of the EMC Celerra File Server system with additional
features and capabilities.
Figure 103 below pictures EMC’s Celerra.
Figure 103. EMC’s Celerra

Source: EMC

Celerra is a unique Celerra is a unique approach to NAS in that it is not a complete NAS appliance. It is
approach to NAS in that a series of NAS servers (thin file servers) without the traditional disk integration.
it is not a complete NAS
appliance. It is a series
Celerra attaches a Symmetrix disk subsystem (EMC’s traditional disk subsystem) on
of NAS servers without its back end, potentially forming a type of SAN architecture behind Celerra (we
the traditional disk would refer to this design as a NAS/SAN hybrid). In other words, Celerra provides
integration. a gateway for clients to directly access files over the LAN (typically Symmetrix disk
subsystems are directly attached to application servers or SANs, thereby forcing
clients to traverse application servers in order to retrieve files). Figure 104 below
illustrates how a Celerra-based NAS architecture might look.

106
The SAN Book III – October 22, 2001

Figure 104. Traditional Celerra Architecture


Tape
Library

Key Considerations
1)Relatively low cost of
Client 1
Server 1
management
Symmetrix 1
2)Fault tolerant
Client 2
3)Data transmission speeds
LAN Celerra 1
are somewhat constrained
by the LAN
Symmetrix 2
4)Heterogeneous
Client 3
connectivity
5)Scalable at the cost of
losing its simplicity
Client 4 Celerra 2 Symmetrix 3

Source: Salomon Smith Barney

Midrange NAS
Mid-range NAS devices According to Dataquest, midrange NAS devices typically have SCSI drives
support a number of (although ATA ones are starting to gain traction), a mixture of software and
drives and RAID, and
offer RAS (reliability,
hardware RAID controllers, and more storage management software functionality
availability, and such as clustering. Pricing ranges from $5,000 to $100,000 with capacities scaling
serviceability) features. to less than 2 TB. Other definitions suggest that these products typically have no
single point of failure and allow field-replaceable units to be swapped while the
system continues to operate (referred to as “hot swapping”). They incorporate some
performance features, such as faster write for RAID configurations, complex
caching algorithms, and/or dedicated data paths.
These systems can be scaled many ways, allowing easy growth in performance,
resiliency, and connectivity. Integrated backup is often offered as an option, and
administration of all network-connected storage can typically be performed from
one central location.
Examples of midrange NAS devices include Procom’s NetFORCE 1700, Network
Appliance’s F740, and Compaq’s TaskSmart N2400. In December, 2000, EMC
introduced its CLARiiON IP4700, a mid-tier NAS product (which can also scale to
the high end) with fault tolerant hardware, a wide range of network connections
including Gigabit Ethernet, and internal failover (processors, power supplies,
pathways). The IP4700 is available in a rack-mountable configuration and supports
from ten to 100 disks utilizing building block arrays. The array utilizes either 18- or
36-GB disks with a total capacity of up to 3.6 TB. EMC noted that the average
setup time has been ten minutes. The operating system is based on software from
CrosStor, which EMC acquired on November 1, 2000.
The Quantum Snap Server ES12 has 1 TB of raw capacity in a 3U rack-mountable
form factor with hot-swappable drives and hot-swappable, redundant power supplies
for under $25,000. Quantum recently acquired the Connex NAS business from
Western Digital to add additional enterprise-class functionality to its product

107
The SAN Book III – October 22, 2001

offering which could move it up the value chain (to the high-end market). Figure
105 below pictures EMC’s CLARiiON IP4700, which can also scale to the high end.
Figure 105. EMC’s CLARiiON FC4700

Source: EMC

Common applications include additional capacity for small and remote offices as
well as file serving, desktop backup, software distribution, and temporary “scratch”
space for IS administrators in large organizations. Common departmental and
enterprise NAS applications include capacity expansion and high-capacity, high-
performance file sharing.

Entry NAS
Dataquest defines entry- Dataquest defines Entry NAS devices as having one to four ATA hard disk drives.
level NAS devices as The low-end of Entry will typically contain one or two drives with either no RAID
having one or two hard
disk drives and no
or RAID 1 while the high-end of Entry will have four drives with RAID 5 support.
expandability. Pricing is usually less than $5,000 with raw capacity up to 300 GB. Other
definitions suggest that these products may be resistant to the most common failures
(such as disk drive, power supply, and cooling system) but are still vulnerable to less
common component failures and usually must be turned off for planned
maintenance.
Connectivity is sometimes limited to one topology (Ethernet, for example).
Common applications include additional capacity for small office/home office, small
businesses, and remote or branch offices, as well as file serving, desktop backup,
software distribution, and temporary “scratch” space for IT administrators in large
organizations. There is much evangelizing that could be done in this nascent
market, as we believe much of the target customer base is not aware of these
products and the benefits they offer. Examples of low-end NAS devices include
Quantum’s Snap Server and Maxtor’s MaxAttach Desktop family.
Figure 106 below pictures Quantum’s Snap 4100.

108
The SAN Book III – October 22, 2001

Figure 106. Quantum’s Snap 4100

Source: Quantum.

Maxtor’s product also In July, 2001, Maxtor announced general availability for its 400 GB, 1U high
has dual 10/100 Ethernet Maxtor MaxAttach NAS 4300 File Server. The MaxAttach 4300 is based on the
ports, gigabit Ethernet
(320 GB), and a SCSI
Windows operating system, and we believe positions the company strongly at the
port for local tape high end of the entry group. Maxtor’s product also has dual 10/100 Ethernet ports,
backup. gigabit Ethernet (320 GB), and a SCSI port for local tape backup. We note that
most NAS filers in this product category do not have Gigabit Ethernet connectivity
nor do they offer an external SCSI port for tape backup. In addition, Maxtor’s
MaxAttach NAS file servers are compatible with backup software from VERITAS,
Legato, and Computer Associates, several device-management tools and support
RAID 0, 1, and 5.
Figure 107 below features the MaxAttach 4100.
Figure 107. Maxtor’s MaxAttach 4100

10/100 Disk
Soft
Ethernet activity
power
monitor

Power UPS port


Dual 10/100 SCSI GigE
Ethernet ports port port

Source: Maxtor and Salomon Smith Barney

109
The SAN Book III – October 22, 2001

Veritas NAS Software


In June, 2001, Veritas announced ServPoint Appliance, a new product that
represents an innovative approach to the NAS market. Instead of developing its own
NAS appliance complete with its own hardware, Veritas has developed software that
will turn a general-purpose server into a file server. The addition of storage to the
file server would result in a NAS-like configuration. The ServPoint Appliance
Software for NAS leverages Veritas’s extensive experience with file systems. We
believe ServPoint provides server vendors with a simple solution to offer NAS
functionality to its customers without the need to develop a new NAS device from
the ground up. It provides Veritas with access to the NAS market without the need
to develop a hardware product.

NAS Versus Traditional Servers


NAS appliances can also be employed in lieu of adding storage to an existing
general-purpose server. Adding a NAS appliance increases processing power to the
network, offloading file-serving duties from the general-purpose server and thereby
improving overall computing performance. Since disks and subsystems do not have
the capability to process the file system, adding a disk or subsystem to a general-
purpose server adds to its burden.
The main difference In their simplest form, NAS appliances are thin servers optimized for file serving
between a general- with plenty of built-in storage and the ability to easily expand and add more storage
purpose server and a
NAS appliance is that the
(most general-purpose servers have built-in storage, too.) The main difference
general-purpose server between a general-purpose server and a NAS appliance is that the general-purpose
can be used for server can be used for applications beyond storing and serving files. These might
applications beyond include serving print jobs, e-mail, applications, databases, and Web pages. The
storing and serving files.
general-purpose server can perform more functions because it has more processing
power, additional interconnects for devices such as monitors, and application
software capable of performing tasks such as transaction and database processing —
all of which take substantial processing power.
Many NAS appliances Many of the components in NAS appliances — especially in the entry level and
use off-the-shelf PC mid-range — are off-the-shelf PC components, as illustrated in Figure 108 below.
components (CPUs,
RAM, and disk drives).
These additional components include microprocessors and related chipsets, memory,
and hard disk drives.

110
The SAN Book III – October 22, 2001

Figure 108. A General Purpose Server Is More Complex — and Expensive — Than a NAS Appliance

PC or General-purpose Server NAS Appliance


System Board yes yes, often less powerful
Enclosure yes yes
Hard Disk Drives yes yes
Server OS License Windows NT proprietary
Client OS Licenses Windows NT proprietary
Power Supply yes yes
Video board none
Ethernet Connection board PCI card
Floppy Disk Drive yes high-end
Monitor yes no
Keyboard yes no
Mouse yes no
Source: Maxtor Corp. (December, 1999), and Salomon Smith Barney

Adding Storage Through NAS


To add storage to a general-purpose server, the server must be shut down and taken
offline. The new storage is then installed, and the server is brought back online.
IDC estimates this process typically takes at least eight hours and often longer.
Some NAS vendors believe the savings on labor to install a NAS device, combined
with a shortage of skilled IT professionals and the benefits of not taking a server
offline, will result in NAS appliances being used to add storage to a general-purpose
server.
Figure 109 below compares the cost of three alternatives for adding storage to a
network: 1) adding hard drives to an existing NT server, 2) adding a Quantum NAS
appliance, and 3) adding an NT server.
Figure 109. Adding More Storage: Quantum’s Entry NAS Appliances Versus an NT Server or Hard Drive
Addition

Add SCSI Drives to an Add a Snap Server Add an NT Server


NT Server to the Network to the Network
Installed Cost $400-$1,999 $499-$4,499 $3000-$10,000
(Parts & Labor)

Required Expertise Professional IT Basic PC Literacy Professional IT


Knowledge Knowledge

Installation Time 2 to 10 hours 5 minutes 1 to 2 days

Downtime During Server is offline None Server is offline hours to


Installation days
Ongoing Management Significant; frequent Minimal Significant; frequent
& Availability weekly reboots weekly reboots

Source: Quantum and Salomon Smith Barney

111
The SAN Book III – October 22, 2001

RAID Is No Good Without Core


Software
➤ Without the secret software sauce of file system and volume
management technologies, RAID might not be of any use at all.

What Does Veritas Have to Do with This?


Veritas has developed the industry’s leading core technology that interconnects
servers with storage subsystems. While HBAs, cables and switches, physically
interconnect servers and storage to provide a next generation architecture, Veritas’s
file system and volume management technologies (which it refers to as its
foundation software) create the intelligence to logically interconnect and leverage
server and storage technologies (i.e., they enable data management). Its no wonder
that the leading server (Sun) and storage (EMC) vendors are old friends of Veritas.
It’s no wonder that the The question is: will they continue to be friends as each peers into the others’
leading server (Sun) and markets and can no longer resist the opportunity for growth, account control, and
storage (EMC) vendors gross margin expansion? As we see it: No. Increasingly, they are all becoming
are old friends of Veritas.
fierce competitors.
Note: We go into greater details of the storage software market in the section titled
“Software is the Magic Word.”
Below in Figure 110, we have illustrated the location of the file system and volume
management within a Direct Attached Storage (DAS) architecture.
Figure 110. File System and Volume Management Architecture

RAID

Disk
OS
Disk
FS
Disk
VM

OS = Operating System FS = File System


VM = Volume Management
Source: Salomon Smith Barney

112
The SAN Book III – October 22, 2001

What Do We Find So Cool About Veritas?


Veritas has displaced the traditional operating system approaches of managing data.
In other words, Veritas has introduced third-party file system and volume
management technology that replaces the operating system’s embedded file system
and volume management functionality. Since Veritas has created superior
independent technology, end users have opted to pay extra for a Veritas add-on in
order to get better and more reliable performance. Sun Solaris was the first
operating system Veritas penetrated. More recently, Microsoft, IBM, and Hewlett-
Packard have partnered with Veritas to invite its file system and volume
management technology into the various operating systems in order to compete
more effectively with the Sun/Veritas combination. Note: Microsoft only uses
Veritas’s Volume Management technology.
Being the foundation technology for NT and all major UNIX operating systems
makes Veritas’s technology the industry standard software platform to manage data.
So, with Veritas on its way up the software stack and trying to make hardware a
commodity, are server and storage vendors planning to punch back? You bet!
Although we cover most of this debate in our section “Software is the Magic Word,”
it is important to stop here and look at the interrelationships of servers, file and file
system management, block and volume management, RAID, external storage and
(ultimately, maybe) virtualization.
We start by building up from blocks, to volume management, to files, to file system
management, and then a quick stop to virtualization.

What Is a Block?
Disks are subdivided A block represents physical disk space. Blocks are often thought of as the lowest
into blocks, which are common denominator of storage (i.e., data). Disks are subdivided into blocks,
numbered and laid out
onto the round disk
which are numbered and laid out onto the round disk platters in concentric circles,
platters in concentric similar to the lines on a phonograph album. Blocks are grouped into tracks, sectors
circles, similar to the and cylinders, but this is beyond the scope of our report. Blocks are the lowest
lines on a phonograph
common denominator of data.
album.
Blocks are stable locations of physical disk storage. Blocks are written to (or filled
with data) in order. In other words, first block 0 is filled, then block 1, then block 2,
etc. This simplicity is part of the reason block storage is often preferred. With
simplicity comes greater reliability, better performance, lower CPU utilization and
an easy, common, standard unit of data.
However, using multiple blocks, as storage does, creates the need to intelligently
manage these blocks through grouping and organization (i.e., volume management
software).
Below in Figure 111, we illustrate the concept of block storage, which is physical
storage. One important thing to note is that multiple blocks can be stored on one
disk (each cylinder below represent one disk, multiple disks form a storage
subsystem). Further, block-based storage groups blocks together, one next to the
other, until a disk is full. This means that data is saved in the order it is created
(conceptually). Therefore, when you go to retrieve data (composed of multiple

113
The SAN Book III – October 22, 2001

blocks), with block-based storage they can all be next to each other (conceptually,
although not always), which would make them easy to find. In other words, it’s
easier to find a box of needles in a haystack, than a couple of hundred individual
needles scattered about (as is sometimes the case with file-based storage).
Figure 111. Block-Based Storage

B Blocks

Disk 1 Disk 2 Disk 3

A
B
C

Subsystem
Source: Salomon Smith Barney

Volume Management
Volume management is Volume management is the management of block data. While blocks are the lowest
the management of block common denominator of data storage, they can be organized into logical groups via
data.
volume management. In other words, volume management technology operates at
the volume level, a step above the block level. Think of it as similar to school. The
twelfth grade is not a person (a block), it’s a way to more logically arrange people
(data blocks) and manage them (in this case to teach) in a more unified and
organized way. Think of blocks as people and volumes as grades. Below in Figure
112 we depict blocks aggregating into volumes.

114
The SAN Book III – October 22, 2001

Figure 112. The Hierarchy of Data Storage — Volume Management

Block Block Block Block Storage

Connectivity

Volume Volume Volume Volume Server

Source: Salomon Smith Barney

Volume management is more flexible than simple block-based storage. Therefore, it


is needed on servers, and sometimes on storage subsystems, for cohesive data
management (as illustrated below in Figure 113). Note: In the diagram below, the
blocks and volumes are located on different disks within the same storage
subsystem.
Figure 113. Volume Management Provides Greater Manageability Than Simple Block Storage
Volumes

Disk 1 Disk 2 Disk 3

A B C

Subsystem
Source: Salomon Smith Barney

115
The SAN Book III – October 22, 2001

Typically, servers send blocks (via volume management technology) to storage


subsystems (RAID), in order to write and read data from disks.

EMC Uses Veritas’s Volume Management and Has Its


Own Volume Manager?
EMC has worked closely
EMC is one of the few (and arguably the only) storage subsystem vendors to
with Veritas to create an integrate its own volume management technology and tightly link it to Veritas’s
enhanced Veritas volume volume manager and file system technology. This has enabled EMC to unlock and
management edition for
create a lot of valuable management features, such as storage-based replication,
EMC’s Symmetrix.
which is an important piece of EMC’s flagship SRDF Disaster Recovery (DR) and
TimeFinder technologies.
EMC has worked closely with Veritas to create an enhanced Veritas volume
management edition for EMC’s Symmetrix. By tightly integrating EMC’s storage
subsystems and volume management technologies with Veritas’s volume manager,
EMC has been able to take advantage of Veritas’s sophistication and provide the
industry’s leading storage solution.
Note: According to sources, (to date) no other storage company has integrated as
tightly with Veritas’s Volume Manager as EMC.

What Is a File?
Think of files as Files are collections of blocks, but are more abstract than physical blocks. Think of
abstractions of blocks. files as abstractions of blocks. Files are not bound to any physical location. Files
are given names and organized in a non-linear format. In other words, while block
zero, block one, and block two are typically laid out one after the other, file X, file
Y, and file Z could all be on different volumes and/or on separate disks and/or
storage subsystems.
Below in Figure 114, we have illustrated how files might be logically grouped and
stored.

116
The SAN Book III – October 22, 2001

Figure 114. File-Based Storage

Volumes

A File X

B File Y Files

File Z
C

Disk 1 Disk 2 Disk 3

A B C

Subsystem

Source: Salomon Smith Barney

File Systems
Since files are abstract, Since files are abstract, they create a more virtualized environment that can be more
they create a more flexible than a physical environment. Put another way, blocks can be renamed into
virtualized environment
that can be more flexible
files and grouped logically (the same concept applies to volumes). Consequently,
than a physical file system software creates abstractions of blocks to improve data management (as
environment. illustrated below in Figure 115). For example, file one might be blocks 17–22 and
file two could be blocks two, nine to 14, and 23. Thought of another way, blocks
one to four could be file 17 and block five could be file two.

117
The SAN Book III – October 22, 2001

Figure 115. The Hierarchy of Data Storage — File Systems

Block Block Block Block Storage

Connectivity

Volume Volume Volume Volume

Server

File System File System File System File System

Source: Salomon Smith Barney

So why make things so confusing with file system software and logically, rather
than physically, group data? Imagine if you physically stored data A, B, and C as
we did in the block storage diagram (Figure 111). Then we deleted data B. Then we
stored data D, E, and F. According to block storage layouts, block D would be laid
out after block C, leaving the space allocated to block B empty. This creates “holes”
of empty disk space which must then be logically filled with data. Unlike a file
system, a volume of block storage has no notion of “free,” “empty,” or “hole.” A
block is a block is a block, and all blocks exist on the volume continuously. One of
the things a file system adds is the notion of blocks being in use or free (a database
has a similar notion internal to the database). Bottom line, of course there is a
reason for the madness.
High performance file The better the volume management and file system technologies, the better the
system software can access to data. High performance file system software can reduce search times by
reduce search times.
providing quality data organization, including look ahead, and better utilize disk
capacity. This is similar to having a really good office filing system, rather than
simply tossing papers onto a pile and then sifting through the entire pile when a
single piece of data is needed. If it’s stored, organized, and managed optimally, it’s
easier to find and use. It takes less brain power to find it (processor utilization) and
it can be found faster (performance).
Think of file systems similar to Microsoft Explorer. Imagine you create three
directories (similar to volumes): A, B, and C. As employee data is organized it is
grouped into directories A , B, and C. Then Homer, an employee from group A
transfers to group B. Simply click on, drag and drop Homer’s data from directory
(volume) A to B. While the file layout has changed (i.e., Homer’s data is now
accessed under the B directory, instead of the A directory) the data (or blocks) can
remain in the same physical location.

118
The SAN Book III – October 22, 2001

Why Are Most Using Veritas’s Core Technology?


Similar to good volume management technology, quality file system technology can
cut through a lot of potential headaches and significantly improve a storage
environment. According to Veritas, a good file system has:
1 high integrity — data remains the same between the process of storing and
retrieving it,
2 availability — being able to retrieve data with ease, and
3 performance — retrieving data quickly through quality organization.
We have outlined two of Veritas’s file system’s important advantages. (Note:
Many of these attributes can also be found in file systems by EMC, Network
Appliance, and others):
1. Journaling
Journaling is used by file systems to balance increasing performance without
compromising data integrity. Journaling can have a significant impact on data
availability.
Here’s the typical problem: System writes are stored in the system’s memory buffer
cache. Therefore, if the system fails before data is written from the cache buffer to
disk, data is lost. In the middle of this process, the file system data is constantly
moving from the cache buffer to disk and vice versa. Therefore, a secondary impact
of a buffer cache failure is that when the failure is circumvented, the failed part is
swapped out or the system is rebooted, the system has to figure out where it left off.
In other words, did it write the data to disk? Did the entire transfer make it before
the failure? This diagnostics check takes time, thereby, delaying the system’s
availability and potentially compromising its reliability. (It’s similar to when a PC
crashes and then has to reboot. The PC takes longer to reboot because it’s trying to
figure out where it left off and if it’s in working order.)
While most file systems While most file systems examine every data block during this process, Veritas’
examine every data advanced journaling system simply looks at the ones recently in use or changed.
block during this
process, Veritas simply
This significantly increases performance upon failure. Veritas also increases data
looks at the ones availability and integrity through other features.
recently in use or
changed. 2. Disk Allocation
Disk allocation is one of the main challenges in data storage. This is the way data is
laid out onto disks. Therefore, the method of block allocation can be a key
differentiator between vendors’ file systems. This is one of Veritas’s key
advantages. According to Veritas there are four major types of disk allocation:
contiguous, linked, indexed, and extent-based. While Veritas uses the more
advanced extent based allocation (which modifies and combines the contiguous and
indexed allocation methods), the other three have been around longer and are,
therefore, more broadly used by legacy operating systems.
Based on Veritas’s input, we have outlined the various allocation methods below:

119
The SAN Book III – October 22, 2001

Contiguous Block Allocation


This method allocates disk space on a contiguous basis. Since data is laid out onto
disks contiguously, it is easier to find. For example, once I get the first word of a
paragraph, the rest are right next to it in order and in the same place. The drawbacks
are when data is deleted, causing “holes,” or when disk space runs out.
Contiguous block allocation must be used in environments without a file system.
IBM’s VM/CMS and Apple use this method.

Linked Block Allocation


This method leaves “pointers” at the end of each disk block to point to where the
next data block is located. While this architecture creates a more flexible
environment and addresses the issue of holes, it can significantly and negatively
impact performance.
MS-DOS, OS/2, Network Appliance, and EMC use this method.

Indexed Block Allocation


This method is similar to linked block allocation in that is uses pointers; however, it
stores the pointers in the file system, rather than with the data blocks on the disks
themselves, thereby freeing up disk capacity and enabling the access of data with
fewer disk head movements. This provides a performance boost over linked block
allocation, but it still increases overhead which negatively impacts performance
relative to contiguous block allocation.
UNIX uses this method.

Extent-Based Allocation
This method takes the Contiguous and Index methods a step further by combining
them and adding features to increase performance (the latter of which is its key
advantage) while decreasing disk capacity overhead and incorporating the use of
pointers (but cuts down on their use relative to the Linked and Index methods) and
disk capacity from holes.
Veritas and EMC’s Celerra use this method.

NAS Changes the File System Rules


Network Appliance’s file While Veritas’s file system is usually located on the server (host-based), Network
system is located on a Appliance’s file system is located on a NAS appliance. This architecture offloads
NAS appliance and
offloads CPU demand to
CPU demand from more expensive application servers to less expensive file servers
less expensive file or NAS appliances (which contain file server functionality). Note: Veritas and EMC
servers or NAS can also locate their file systems on a NAS appliance or file server and, in fact, are
appliances. aggressively expanding into this market.
In the figures below we illustrate the location of the file system in Direct Attached
Storage (DAS), Storage Area Network (SAN), and Network Attached Storage
(NAS) architectures. Note: While the location of the file system can be moved
around, the core technology is largely the same.

120
The SAN Book III – October 22, 2001

Figure 116. File System DAS Architecture

Tape
Library

Client 1
FS
Subsystem 1
Server 1
Client 2
LAN
FS Subsystem 2
Client 3 Server 2

Client 4 FS Subsystem 3
Server 3

Source: Salomon Smith Barney

Figure 117. File System SAN Architecture

Tape
Library

Client 1 FS
Server 1
Subsystem 1
Client 2
LAN SAN
FS
Client 3 Server 2
Subsystem 2

Client 4 FS
Server 3 Subsystem 3

Source: Salomon Smith Barney

121
The SAN Book III – October 22, 2001

Figure 118. File System NAS Architecture

Client 1

Server 1

Client 2
LAN

Client 3 Server 2

FS
Client 4 NAS
FS Appliance 1

NAS
Appliance 2

Source: Salomon Smith Barney

Architecturally
On the other hand, DAS Architecturally we can look at NAS as connecting to or obviating the need for
or SANs use simpler, application server file systems, which can reduce file system expenses (each copy of
more efficient volume
management (block)
a file system costs money) and more expensive application server CPU utilization
based storage as an (although NAS creates greater overall CPU demand, it offloads the CPU cycles from
interconnect. more expensive application servers onto less expensive file servers or NAS). On the
other hand, DAS or SANs use simpler, more efficient volume management (block)
based storage as an interconnect which is often viewed as more reliable and faster,
particularly for compute of data intensive applications. We have illustrated this in
Figure 119 below.

122
The SAN Book III – October 22, 2001

Figure 119. Various Data Architectures

Application
Server
Application Software
NAS Operating System
File Server
Functionality File System
File System
Volume Management
Volume Management
RAID

Disk

Volume Management

RAID

Disk

Storage
Subsystem
Source: Salomon Smith Barney

Storage subsystems can Note: Storage subsystems can have their own volume management technology.
have their own volume Although it is uncommon, this has been one of EMC’s key differentiators on its high-
management technology.
Although it is
end subsystems.
uncommon, this has
been one of EMC’s key
Network Appliance’s File System
differentiators on its One of Network Appliance’s most distinctive advantages over traditional server-
high-end subsystems. attached storage, as well as its NAS competition, is its sophisticated file system,
which is strategically located on its NAS appliance’s thin server. In conventional
local or direct attached storage and block-level SAN implementations, the file
system is colocated with the operating system/application server — separate from
the RAID controller. In this approach, file systems send blocks to the RAID
controller, which lays the data blocks out on the disk array. By separating the file
system from the controller/subsystems, Network Appliance believes that file system
advantages such as security, logical volume and file management, and data
management can be lost (although many would argue the opposite).
Network Appliance’s filers integrate its file system — called Write Anywhere File
Layout (WAFL) — directly with the RAID array, which preserves and enhances
file-mode access advantages. Network Appliance believes benefits for local file-
sharing access include scalability, management, resiliency, failover, and lower total
cost of ownership. To date, Network Appliance filers have been attached to
application servers and clients through TCP/IP utilizing NFS, CIFS, and HTTP
protocols.

123
The SAN Book III – October 22, 2001

Heterogeneous Connectivity
In traditional server-attached storage architectures, storage subsystems typically
cannot share files between platforms. This is partly due to the location of the file
system, which is located on the application server. In effect, the location of a file
system in a traditional storage architecture couples each storage subsystem with the
operating or file system being used by the application server, as illustrated below in
Figure 120.
Figure 120. Storage Subsystem Connectivity by Operating and File System

Mainframe Mainframe

FS

UNIX UNIX

FS

NT NT

FS
Source: Salomon Smith Barney

Being able to provide Being able to provide heterogeneous connectivity between discrepant systems and
heterogeneous technologies is a powerful advantage. For example, would it be nice to store UNIX
connectivity between
discrepant systems and
and NT data onto a single storage subsystem? Network Appliance can due to its file
technologies is a system architecture, but so can EMC using its block-based architecture.
powerful advantage.
Heterogeneous data storage has been one of EMC’s largest competitive advantages
in the past. Recently many other storage subsystem vendors have developed
heterogeneous connectivity, although EMC still supports the broadest number and
versions of operating systems, including the mainframe. Block based storage
subsystems can store UNIX and NT files on the same subsystem and then translate,
if requested, NT data into UNIX data. However, due to its file system architecture
Network Appliance has taken heterogeneous connectivity to a more advanced level.
Block-based subsystems partition disks by operating system (e.g., NT, UNIX). It
can then cache the data and translate it to communicate with other operating
systems. In Figure 121 below, we have illustrated how a storage subsystem stores
data and retrieves it by operating system.

124
The SAN Book III – October 22, 2001

Figure 121. Heterogeneous Storage

Mainframe

FS Subsystem

Mainframe
UNIX

FS UNIX

NT
NT

FS

Source: Salomon Smith Barney

What sets Network What sets Network Appliance’s heterogeneous storage apart from others is that it
Appliance’s actually stores data in a neutral (neither NT nor UNIX) format. This obviates the
heterogeneous storage
apart from others is that
need to “translate” data back and forth, using expensive caching buffers. However,
it actually stores data in it does add overhead at the onset of a save or write command. Below in Figure 122,
a neutral (neither NT nor we have anecdotally illustrated how Network Appliance stores data in a neutral
UNIX) format. format.
Figure 122. Network Appliance Uniquely Stores Data in a Heterogeneous, Neutral Format

UNIX UNIX Neutral

A Neutral A A’

NT OR
NT Neutral

B
B B’

Source: Salomon Smith Barney

About 40% of all Network Appliance systems shipped are installed out of the box to
interconnect multiple computing platforms, typically UNIX and NT. By leveraging
its file-system design, Network Appliance is able to offer heterogeneous
connectivity while maintaining only one file. To offer heterogeneous connectivity,

125
The SAN Book III – October 22, 2001

other storage systems need to maintain multiple copies of the same file, increasing
complexity and requiring additional storage space.
Below in Figure 123 we have illustrated how Network Appliance’s heterogeneous
connectivity might be viewed. Note: Network Appliance does not fully support the
mainframe.
Figure 123. Network Appliance’s Heterogeneous Storage

NetApp
Appliance
UNIX
FS
UNIX &
NT

NT

Source: Salomon Smith Barney

Scalability
Simply snap new disk Files organize data logically, instead of physically (which is done by volume
capacity (i.e., physical mangers). This means that blocks can be any size, regardless of physical limitations.
disk) onto the network
and an intelligent file
Thought of another way, simply snap new disk capacity (i.e. physical disk) onto the
system can seamlessly network and an intelligent file system can seamlessly assimilate that new storage.
assimilate that new With volume management software, a new node must be initialized, which
storage. sometimes takes the environment temporarily offline. Blocks tend to be a fixed size
while files and file systems can be any size.
No Free Lunches
File-based storage also has its downfalls. For example, it requires higher CPU
utilization and is more complex than block-based storage. These complexities can
increase the chances of failure and negatively impact performance. Often times in
computing, simpler is better.

Combining the Best of Both Worlds: EMC’s HighRoad


In December, 2000, EMC introduced software enhancements for its Celerra NAS
server called HighRoad along with a new high-end Celerra. EMC’s Celerra
HighRoad File System Software, which enhances the functionality of the EMC
Celerra, allows files to be served either through Celerra or through the SAN. The
software intelligently determines the profile of the requested data and then optimally
determines the path to best serve the data to the client. The software is installed on
Celerra with agents located on participating servers connected to a SAN. Celerra SE
is a high-end NAS device housing both EMC Celerra and Symmetrix Storage
systems in a single-bay enclosure. CelerraSE scales to 8TB of raw capacity.

126
The SAN Book III – October 22, 2001

EMC’s Celerra and HighRoad Architecture


Celerra and HighRoads offload the file system and data path to 1) reduce cost, and
2) increase performance. Although the downside is that this architecture greatly
increases server-to-storage complexity, some argue that its advantages far outweigh
this factor.

Externalizing the File System Can Save $$$


By offloading By offloading (externalizing) the file system onto a separate file server, file system
(externalizing) the file costs can be reduced. For example, in an environment with 100 application servers,
system onto a separate
file server, cost can be
100 file system licenses are required. On the other hand, one (or a couple of)
significantly reduced. Celerra file server(s) could reduce the number of licenses needed; thereby,
potentially reducing cost.
Veritas and Network Appliance can also offload the file system from application
servers if desired. Some argue that this architecture creates a potential bottleneck in
highly active compute environments. Its also creates a more complex environment
with additional networking, hardware and management requirements, and introduces
a very sensitive single point of failure. In the end, while this architecture might be
optimal in some environments, it certainly does not suit all. EMC is focusing on a
distributed storage management approach that has the flexibility to distribute
intelligence.
Figure 124. Externalizing the File System using EMC’s Celerra

Tape
Library

Server 1
Client 1

Subsystem 1
Client 2
LAN Server 2
SAN

Client 3 Subsystem 2

Server 3
Client 4

Subsystem 3
FS

Source: Salomon Smith Barney

EMC’s Highroads greatly enhances storage environments by:


➤ Greater Connectivity: creating greater Symmetrix connectivity
➤ Being Faster: increases performance by creating multiple paths of delivery; also
one path is Fibre Channel, which is faster than more cumbersome Ethernet
➤ HA Architecture: by creating multiple paths, HighRoads enhances data
availability in case of a path or system failure

127
The SAN Book III – October 22, 2001

➤ Optimal Performance: HighRoads can dynamically determine the most


efficient data path, be it through a file server using files for smaller transfers or
an applications server using bocks for larger data transfers
➤ Reducing Network Traffic: by offloading potential application server-to-
Celerra (file server) data traffic onto a dedicated Fibre Channel storage network,
LAN traffic can be reduced
Figure 125. EMC’s Celerra Without HighRoad

Server 1
Client 1

Client 2 Subsystem 1
LAN Server 2

Client 3

NAS File
Client 4 Server 1
Subsystem 2

Source: Salomon Smith Barney

Figure 126. EMC’s HighRoad Software Enables Greater Symmetrix Connectivity

Server 1
Client 1

Client 2 Subsystem 1
LAN Server 2

Client 3

NAS File
Client 4 Server 1
Subsystem 2

Source: Salomon Smith Barney

By implementing HighRoads in a SAN environment, further efficiencies can be


created by incorporating the benefits of SANs as shown in Figure 127 below.

128
The SAN Book III – October 22, 2001

Figure 127. EMC’s HighRoads Can Leverage SANs for Best-in-Class Performance

Tape
Library

Server 1
Client 1

Subsystem 1
Client 2
LAN Server 2
SAN

Client 3 Subsystem 2

Server 3
Client 4

Subsystem 3
FS

Source: Salomon Smith Barney

Competition...Competition!!!
Its no wonder to us that Network Appliance, EMC, and Veritas are becoming fierce
competitors for data management and file management technology.

Virtual Reality?
Essentially, virtualization Virtualization is the natural evolution in data management. We started with disks
is a hierarchical (blocks), then grouped them into subsystems (volumes), and then organized the data
grouping of volumes that
allow the management of
in the server (file systems). Virtualization is the next logical evolutionary (not
all storage as a common revolutionary) step in data management. Essentially, it’s a hierarchical grouping of
pool. volumes that allows the management of all storage as a common pool. Below in
Figures 128 and 129 we have depicted how this might fit into our block diagram.

129
The SAN Book III – October 22, 2001

Figure 128. Virtualization in a Server-Centric Architecture

Block Block Block Block Storage

Connectivity

Volume Volume Volume Volume

Virtual Layer Virtual Layer Virtual Layer Virtual Layer Server

File System File System File System File System

Source: Salomon Smith Barney

Figure 129. Virtualization in a Network-Centric Architecture

Block Block Block Block Storage

Volume Volume Volume Volume Connectivity

Virtual Layer Virtual Layer Virtual Layer Virtual Layer

File System File System File System File System Server

Source: Salomon Smith Barney

Note: The virtualization architecture could also be storage-centric, meaning that


the virtualization technology would be located within a storage subsystem. Or, it
can be distributed, as is EMC’s view with its AutoIS.
Many startups, such as StorageApps (bought by Hewlett-Packard), trueSAN,
DataDirect, FalconStor, VICOM, StoreAge, and XIOtech (bought by Seagate), and
many existing companies (such as Compaq, EMC, StorageTek and Veritas) have
been developing virtualization software in order to create the ultimate, ubiquitous,
heterogeneous storage pool that can expand exponentially. However, it is important
to keep in mind that while many talk about virtualization as a silver bullet to

130
The SAN Book III – October 22, 2001

ubiquitous data management in the near term, we do not expect to see it in a broadly
usable form for a number of years.
Virtualization introduces Virtual Promise: Unlimited storage capacity on demand. In other words, capacity
the concept of “free” can be added on the fly without taking systems down or having to reconfigure them,
versus “in use” space to
the SAN at the
similar to the way file systems operate. Virtualization introduces the concept of
block/volume level and “free” versus “in use” space to the SAN at the block/volume level and the ability to
the ability to map things map things at a logical level all the way down to a physical level.
at a logical level all the
way down to a physical Below in Figure 130, we have illustrated the way a virtual pool of storage might be
level. interpreted graphically.
Figure 130. Virtual Storage Pool of Unlimited Storage on Demand

SAN Virtual SAN

Virtual Storage
Pool

Source: Salomon Smith Barney

Analogy
Similar to our previous school system analogy, virtualization becomes another layer
or subset.
➤ person = block (disk)
➤ grade = volume (volume management)
➤ school division (primary school, high school, college, etc.) = virtualization
➤ school = file system

131
The SAN Book III – October 22, 2001

This page intentionally left blank.

132
The SAN Book III – October 22, 2001

Software

The Magic Word is Software


Software Applications

133
The SAN Book III – October 22, 2001

The Magic Word Is Software


➤ Software is the secret sauce that makes it all work.

➤ The importance of storage software has increased exponentially, in


our view. As information has gained value, storage environments
have become more complex, and the cost to manage data has
increased.

Software Takes Storage to the Next Level


Software is the glue that Perhaps the most complex aspect of the computer storage evolution for IT managers
that ties a networked has been the actual control and management of its implementation. Data centers
environment together.
may include several CPUs or servers, multiple disk subsystems, and various tape
libraries all distributed across wide geographic areas, manufactured by various
suppliers, and spanning a multitude of computing platforms. Add to that the
doubling of data every year and the fact that budgets are under pressure and it
becomes easy to conclude that the evolution and complexity of computing are at the
risk of becoming unmanageable.
Storage software enables the effective and efficient management of explosive data
growth. Storage software has been one of the key drivers to lowering overall storage
management costs by increasing the productivity per person for administrators.

The Importance of Software


The history of the computer industry leads us to conclude that hardware
development cycles typically get ahead of software development cycles. As an
application area matures, the hardware component tends to diminish in importance,
which is why hardware-based companies are continually striving for the next
breakthrough technology or architecture.
In the initial stages of a software application’s evolution, the application typically
provides minimal functionality (the 1.0 release). Over time, software generally
“matures,” adding functionality. In the later stages of an application’s development,
the software provides most of the value and, since hardware becomes tougher to
differentiate, there is consolidation of hardware vendors.
In addition, hardware vendors sometimes try to commercialize their proprietary
software, which is referred to as “firmware” (software written to work closely with a
specific hardware design). This is often mentally difficult to execute because the
software is tightly integrated with specific proprietary hardware configurations and
commercializing one’s software may enable other hardware vendors to offer a more
competitive solution. The result is that firmware historically has been a tertiary
business, which tends to offer less functionality than can be attained through
independent software providers.
The maturity process of software development can be tracked as follows:
➤ Code is developed;

134
The SAN Book III – October 22, 2001

➤ Bugs or errors are found and fixed;


➤ Code is rewritten to be more efficient, often shortening “path lengths”;
➤ Functionality is added to enhance the application;
➤ The target hardware support is expanded, making the application more open and
heterogeneous; and
➤ Ease of use, documentation, and support are all enhanced.
In summary, we believe software companies typically have a longer lasting value
proposition than hardware companies and require less dominant market shares to
ensure long-term viability.
“But I Don’t Want To Be a Commodity!” — Hardware Vendor
Adding value is the real And who does? This is the battlefront of storage computing. It is an issue of make
competitive versus buy. Does a company delay its entrance into the market by developing in-
differentiator, but not at
the cost of losing market
house, proprietary software, or does it speed up its entrance by OEMing software?
share. The first scenario enables 90% gross margins and locking in the customer to your
solution. The second choice enables a total solution, maintaining account control,
and getting to the market quickly. Which is the more prudent path in our opinion?
Both!
Server Vendors Have Been Leveraging Veritas
Both Hewlett-Packard and IBM have signed up to partner with Veritas to embed its
foundation software (which includes file system and volume management software)
into their operating systems releases. We believe this was done 1) because of the
performance advantages Veritas software provides, and 2) to combat Sun
Microsystems’ partnership with Veritas foundation products, which has provided
Sun with a more competitive offering resulting in greater market share. At the same
time Sun, Hewlett-Packard, and IBM each plan to continue developing their own
proprietary UNIX-based OS.
Through Veritas’s Veritas began its file system and Volume Management (VM) proliferation with Sun
dogged efforts, it now Microsystems. In effect, Veritas provided a better file system and VM than what
has the leading market
position in file system
Sun embedded in its Solaris-based (UNIX) servers. When Sun recognized that not
and VM software. only was Veritas’s Solaris foundation suite better than its own but better than many
of its competitors, Sun welcomed Veritas onto its pricing sheet and became a
reseller of Veritas’s Solaris foundation products. We believe this partnership
furthered Sun’s competitive advantage. It took Hewlett-Packard and IBM years to
succumb to Veritas’s file system and VM expertise. Through Veritas’s dogged
efforts, it now has the leading market position in file system and VM software,
capturing 61% of the UNIX market in 2000 (Figure 131 below).
Note: We combine the Core Storage Management Technology (includes file systems)
and Virtualization (includes volume management) categories since Dataquest
categorizes Veritas’s Volume Manager as a Virtualization product. Also, Legato
does not offer a file system or volume manager product so we do not describe its
partnerships here.

135
The SAN Book III – October 22, 2001

Figure 131. Core Storage Management and Virtualization Software, UNIX platform

1999 2000
Other
Other 9%
EMC EMC
10%
23% 30%

Veritas
Veritas 61%
67%

Source: Dataquest (April, 2001)

Veritas is the fastest-growing independent supplier of storage software. The


company grew its software licensing revenues 65% year over year in 2000. Veritas
supplies its file system and volume management software (referred to as its
Foundation suite) for Sun Solaris, EMC’s Symmetrix (VM only), NT (only volume
management rebranded by Microsoft as Logical Disk Manager, not file system
software), Hewlett-Packard HP/UX (embedded with a “lite” version), and IBM
AIX/Monterey (embedded with a “lite” version; expected in the second half of
2001). Why are so many large companies deferring to Veritas’s file system and
volume management software expertise? Because it’s better to get a piece of the
market than nothing at all. It is also important to note that Veritas has proven time
and again that it can provide more robust file system and volume management
software on a more timely basis.
Further, we believe as more companies, such as IBM, begin to ship embedded
Veritas products, Veritas will continue its market share dominance in this segment
(referred to by Dataquest as Core Storage Management and Virtualization).
On a related note, on August 8, 2001, Sun Microsystems and Hitachi Data Systems
announced a three-year agreement under which they will cross-license current
storage software products and collaborate on future ones. Veritas has relationships
with both of these vendors.

The Attraction to Software


The business model for a The business model for a software company is very seductive. Figure 132 below
software company — lists several generic business models that clearly illustrate the financial power of a
including gross margins
of 88%–90% — is very
well-run software company in today’s market.
seductive.

136
The SAN Book III – October 22, 2001

Figure 132. Financial Power Comparison


Computer
AVERAGES Software Networking Leadership H/W CEM/EMS
GM 88.0% 54.0% 50.0% 9.0%
Oper Mgn 30.0% 20.0% 20.0% 5.5%
Net Margins 20.0% 15.0% 14.0% 3.5%

ROE 30.0% 29.0% 24.0% 13.1%


ROA 30.0% 20.0% 13.0% 7.0%

PE/Growth Rate 3.0 2.2 1.5 1.0


Source: Company reports and Salomon Smith Barney

A Competitive and Consolidating Market


With such robust market projections, such an attractive business model, and the
increasing importance of management, it is easy to understand why the storage
software market is so fiercely competitive. Many startup companies have found the
environment too daunting to go it alone. This has led to a flurry of acquisitions by
larger companies with greater resources.
The figures below highlight acquisitions by BMC, Computer Associates, EMC,
IBM, Legato, Network Appliance, and Veritas.
Figure 133. BMC Software Acquisitions
Date Date
Announced Completed Name Type Value at Announcement Product
8-Mar-99 14-Apr-99 New Purchase Share purchase and tender Enterprise management
Dimension agreement to acquire New software; Ouptput
Software Dimension Software, Ltd. for management software
$52.50 per share; Approx.
value of $650M
21-May-97 21-May-97 Datatools Inc. Purchase $60M cash SQL Backtrack, DB
Recovery solutions
Source: BMC Press Releases; Bloomberg

Figure 134. Computer Associates Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
14-Feb-00 10-Apr-00 Sterling Software Stock Purchase; $4B stock-for-stock; exchange E-business software
Inc. purchase accounting 0.5634 shares of CA stock for
each outstanding Sterling share

29-Mar-99 29-Jun-99 Platinum Cash purchase; Computer Associates to pay


Technology purchase accounting $29.25 cash per share of
International Platinum Technology; approx.
$3.5B
27-Oct-96 2-Dec-96 Cheyenne Cash purchase; $1,175M - $27.50 per share
Software purchase accounting

Source: Computer Associates Press Releases; Bloomberg

137
The SAN Book III – October 22, 2001

Figure 135. EMC Corporation Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
20-Sep-01 20-Sep-01 Luminate Cash purchase $50M Performance
monitoring
11-Apr-01 11-Apr-01 FilePool NV Cash purchase $50M Belgian software
company
1-Nov-00 1-Nov-00 CrossStor Stock pooling $300M Networked storage
Software systems software
16-Aug-00 16-Aug-00 Avalon Purchase Undisclosed Rich media archive
Consulting management software
14-Jan-00 14-Jan-00 Terascape Undisclosed $50M cash Software
Software Inc.
21-Dec-99 28-Jan-00 Softworks Inc Undisclosed $192M cash Corp. information
software
9-Aug-99 13-Oct-99 Data General Pooling of interest; $1.1B as of Aug 6; EMC will AViiON software
Corp. accretive to earnings issue 0.3262 of a share of
EMC common stock for each
share of Data General
common stock
10-Aug-98 10-Aug-98 Conley Corp. Undisclosed Undisclosed amount High-availability, high-
performance storage
management software

Source: EMC Press Releases; Bloomberg

Figure 136. IBM Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
14-Dec-99 14-Dec-99 Mercury Cash purchase $23.5 million Software that
Computer's enables file sharing
Storage division on a network
31-Jan-96 4-Mar-96 Tivoli Systems Cash $743 million Tools that manage
corporate computer
networks

Source: IBM Press Releases, Bloomberg

138
The SAN Book III – October 22, 2001

Figure 137. Legato Systems Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
11-Jul-01 20-Jul-01 SCH Cash $12.5M Media life cycle
Technologies management, and tape
device and library sharing
28-Jan-99 1-Apr-99 Intelliguard Purchase accounting; Legato will issue 720,000 shares of Celestra Architecture;
Software, Inc. acquisition accretive to stock and the cash equivalent of Network Data Management
earnings 180,000 share of Legato stock at the Protocol (NDMP)
closing price on the day preceding the
completion of the transaction for all of
the outstanding stock of Intelliguard
Software. Transaction value was
$52M as of Jan. 27
18-Nov-99 Terminated Ontrack Data Purchase accounting; Legato will issue a combination of Data recovery services and
24-Jan-00 International accretive to 2000 earnings approx. 1.485M shares of Legato products
stock and approx. $20M in cash for all
of the outstanding stock of Ontrack
Corp. Approx value of $134M as of
Nov. 17
7-Jun-99 2-Aug-99 Vinca Corp. Purchase accounting; Combination of stock and cash valued StandbyServer family; High
accretive to 1999 earnings at approx. $94M availability data protection
software
26-Oct-98 19-Apr-99 Qualix Group Inc. Pooling of interest; tax-free Legato will issue 1.721M share of Application service level
D/B/A FullTime reorganization stock in exchange for all the stock software solutions
Software Inc. and options of FullTime Software.
Approx. value $69.4M
30-Jul-98 6-Aug-98 Software Moguls. Pooling of interest; tax-free Legato to issue 250,000 shares of its SM-arch; backup-retrieval
Inc. reorganization stock in exchange for all for all the products
stock of Software Moguls. Approx.
value $10.2M

Source: Legato Press Releases; Bloomberg

Figure 138. Network Appliance Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
5-Sep-00 5-Sep-00 WebManage Stock purchase $75M Content management
software
13-Jun-00 13-Jun-00 Orca Systems Stock purchase $49M Virutal Interface (VI)
software
17-Mar-97 17-Mar-97 Internet Stock puchase $7M Internet proxy caching
Middleware software

Source: Network Appliance Press Releases, Bloomberg

139
The SAN Book III – October 22, 2001

Figure 139. Veritas Acquisitions


Date Date
Announced Completed Name Type Value at Announcement Product
15-Feb-01 15-Feb-01 Prassi Europe SAS Cash purchase Undisclosed CD and DVD-mastering
technology
29-Mar-00 30-Sept Seagate Technology Management buyout. Investor group and Seagate None. Financial
approx. Seagate to become private management purchase Seagate's transaction.
operating businesses for $2B.
Then Veritas acquires Seagate,
including all of VRTS shares then
held by Seagate plus investment
security investment in other
companies and cash
17-Aug-99 17-Aug-99 ClusterX, part of NuView, Purchase accounting; cash $67.9M in cash and stock ClusterX
Inc. & stock
1-Sep-98 1-Jun-99 TeleBackup Systems Inc. Amalgamation under 1.7M shares of VRTS, entire deal TSInfoPRO
Canadian law; tax-free; valued at $86M
pooling of interests
5-Oct-98 28-May-99 Seagate Software's Purchase accounting; tax- 33M shares of SEG (for the Storage management
Network Storage free; merger software business valued at software for UNIX and
Management Group $1.6B); VRTS will assume NSMG Windows NT platforms;
employee options. Total approx. Netbackup Professional
value $3.1B
18-Feb-99 3-Feb-99 OpenVision Australia Pty. Subsidiary formed; Undisclosed amount Distribution
Ltd. purchase accounting; non-
dilutive (purchase)
10-Feb-99 8-Feb-99 Frontier Software Wholly owned subsidiary; Undisclosed amount Development Center
Development (India); purchase accounting; non
Private Limited (Pune, dilutive
India operations)
18-May-98 15-May-98 Windward Technologies Purchase accounting; cash Undisclosed amount Windward Lookout failure
Inc. prediction software
13-Jan-97 24-Apr-97 OpenVision Technologies Pooling of interests; tax- 7.5M shares of VRTS. Entire deal Business application
free; merger valued at $126.6M management solutions;
HA, UNIX backup,
Exchange
28-Mar-96 Advanced Computing Purchase accounting; cash $3.5M cash plus royalties, capped Network Removable Media
Systems Inc. and royalties at $5.5M Library (VML Media Lib)

12-Apr-95 10-Apr-95 Tidalwave Technologies Pooling of interests; 117,374 shares of VRTS; FirstWatch failover
Inc. exchange of stock VERITAS assumes a warrant software; HA
which entitles the holder to acquire
an additional 27,010 shares of
VRTS common stock

Source: Veritas Press Releases; Bloomberg

The advantages of a The advantages of a “total solutions” product offering have helped drive
“total solutions” product consolidation in the storage software market. Rather than offering a niche set of
offering have helped
drive the consolidation
software products and competing with larger competitors with a complete software
in the storage software product portfolio, both acquirers and acquirees see the wisdom in merging, thereby
market. increasing the combined companies’ potential addressable market by offering a
more complete total solutions package.
We estimate that 25%– Even with all the consolidation that has already occurred, we believe the storage
30% of the storage software market still has many companies left that will go out of business or will be
software market remains
“white box.”
acquired. We estimate that 15%–20% of the storage software market remains
“white box” (i.e., in the hands of small companies).

Storage Software Trends


In the past, storage software has been somewhat interesting, offering backup,
recovery, device administration, media library management, HSM, file system and
Volume Management (VM) software. But it has been the recent momentum of data
replication, SAN/Device Administration, and SRM software that have brought
mainframe and open-systems computing to a more level playing field. We believe
these budding segments will not only offer significant growth well into the future,

140
The SAN Book III – October 22, 2001

but will increase the value of the other more traditional storage software product
areas.
It has been the recent For example, the development and implementation of SRM and SAN/Device
momentum of data Administration (which are two pieces of the same function; please refer to the
replication, SAN/Device
Administration and SRM
“Software Applications” section for more details on SRM) have enabled
software that have backup/recovery networking to be done more manageably. When systems go on a
brought mainframe and network they often become more complex and harder to manage due to the inclusion
open-systems computing
of many more variables. With SRM and SAN/Device Administration, the
to a more level playing
field. environment can be viewed and altered more easily. SRM will bring the
architecture up into a format that is easy to read and understand and SAN/Device
Administration will alter the environment to make the appropriate policy changes
desired. Therefore, SRM and SAN/Device Administration help generate more
demand for Backup/Recovery Networking (open systems) versus Backup/Recovery
Host (mainframe).
Figure 140 below outlines the storage software market segments in 1998 and 2005
based on Dataquest’s projections and categories. Going back to 1998 might seem
like a bit much, but it helps accentuate the trends. Note the increase in the size of
the market: from $2.9 billion (1998) to $16.7 billion (2005) in seven years,
representing a 29% CAGR between 1998 and 2005. Also note that SRM moves to
10% from 4% of the total market; Virtualization to 7% from 2%; and Replication to
20% from 15%. Note: Dataquest revenue calculations include new license sales but
not revenues from maintenance and support services.
Figure 140. Storage Software Trends

1998 2005
$2.9 Billion $16.7 Billion

Virtualization Storage Resource Media & Library


Core Storage Storage Resource Virtualization
Software Mangement Management
Management Mangement Software
2% 10% 3%
Technology 4% Media & Library 7%
8% Management
Core Storage SAN & Device
8%
Data Replication Management Administration
15% SAN & Device Technology 10%
Administration 16%
5%
HSM / Archive Data Replication
9% 20% Backup /
Recovery Network
Backup / Backup / HSM / Archive
Backup / 25%
Recovery Host Recovery Network 4%
Recovery Host
23% 26%
5%

Source: Dataquest (April, 2001)

NT and UNIX Focus


NT and UNIX are the When looking at specific companies in the storage software market, it is important
fastest-growing to recognize the difference between mainframe (MVS) and open systems (NT and
segments in the storage
software market.
UNIX). In 2000, UNIX storage software revenues surpassed mainframe revenues.
In 2001, Windows NT storage software revenues are projected to also surpass
mainframe revenues (Figure 141 below).

141
The SAN Book III – October 22, 2001

Figure 141. Total Worldwide Storage Software Market Revenues by Platform


CAGR
1998 1999 2000 2001 2002 2003 2004 2005 2000-2005
MVS 1,322.1 1,541.9 1,491.0 1,491.0 1,505.9 1,528.5 1,548.4 1,565.4 1.0%
Unix 745.9 1,306.9 2,094.8 2,792.6 3,766.1 4,967.1 6,407.6 8,265.8 31.6%
Windows NT 478.8 810.9 1,213.8 1,859.5 2,640.5 3,624.9 4,817.8 6,233.6 38.7%
Netware 153.6 207.2 139.9 125.9 107.0 85.6 59.9 54.9 (17.1)%
Proprietary 68.6 86.2 175.4 228.0 285.0 342.0 393.3 432.7 19.8%
Windows 32 61.9 86.8 89.9 98.9 107.8 116.4 124.6 132.0 8.0%
Windows 16 13.6 8.0 3.2 0.8 - - - - NA
VMS 20.2 15.7 18.7 18.7 18.3 17.8 17.1 16.2 (2.8)%
OS/400 8.0 13.1 16.3 18.7 20.6 22.7 24.9 26.2 10.0%
OS/2 6.9 2.5 1.2 0.6 0.3 0.1 - - NA
Macintosh 10.6 8.8 7.2 5.8 4.5 3.4 2.5 1.8 (24.2)%
Total 2,890.2 4,088.0 5,251.4 6,640.5 8,456.0 10,708.5 13,396.1 16,728.6 26.1%

Source: Dataquest (May, 2001)

Most companies are It comes as no surprise to us that most companies are increasing their focus on NT
increasing their focus on and UNIX. In fact, while BMC, Computer Associates, and IBM derived a good
NT and UNIX.
portion of their revenues from the mainframe platform in 2000 (63%, 46%, and
76%, respectively), the up-and-comers such as EMC and Veritas derive the bulk of
their revenues from NT (23% and 32%, respectively) and UNIX (64% and 64%,
respectively). The mainframe storage software market share breakdown for 2000 is
shown in Figure 142 below.
Figure 142. Mainframe Storage Software Market Share (2000)

Other
StorageTek 10%
9%
IBM
BMC
42%
9%

EMC
11%
Computer
Associates
19%

Source: Dataquest (April, 2001). Includes MVS and VMS.

Figure 143 below illustrates the storage software market share on the UNIX
platform in 2000. Veritas has a much larger presence in open systems, on both
UNIX and Windows NT.

142
The SAN Book III – October 22, 2001

Figure 143. UNIX Storage Software Market Share (2000)


Computer Legato
Associates 3%
3% Other
13%
Hewlett-Packard
4%
EMC
BMC 40%
4%

IBM
7%
Veritas
26%

Source: Dataquest (April, 2001)

Figure 144 below illustrates the storage software market share on the Windows NT
platform in 2000.
Figure 144. NT Storage Software Market Share (2000)

Other EMC
22% 26%

Legato
6%
Compaq Veritas
8% Computer 22%
Associates
16%

Source: Dataquest (April, 2001)

Storage Software Projections and Market Share


Software is the glue that Software is the glue that ties together various computing devices and platforms. The
ties together various increasing complexity of data management has made high-level software
computing devices and
platforms.
functionality non-negotiable. As a result, the storage software market has grown
substantially. Dataquest projects that worldwide storage software market revenues
will grow to $16.7 billion in 2005 from $5.3 billion in 2000, representing a 26.1%
CAGR.
In Figure 145 below is the storage software triangle, a layered diagram of the
subsegments which does not exactly fit the order Dataquest uses, but it is the way
we prefer to look at the market.

143
The SAN Book III – October 22, 2001

Figure 145. Total Worldwide Storage Software Market Revenues

2000 2005
Market Size Market Size
Total $5,251M Total $16,729M
CAGR 26.1%

Storage
Resource
$271.3 Mgmt. $1,709.5; 44.5%

$966.9 Data Replication $3,323.2; 28.0%


Applications

$307.1 HSM / Archive $603.8; 14.5%

$1,332.6 Backup / Recovery Network $4,305.9; 26.4%

$697.6 Backup / Recovery Host $768.7; 2.0%

$178.5 Virtualization $1,213.5; 46.7%

NA Clustering NA

$327.2 Media & Library Management $506.6; 9.1%


Core

$378.6 SAN & Device Administration $1,644.7; 34.1%

$791.7 Core $2,652.7; 27.4%

Source: Dataquest (May, 2001) and Salomon Smith Barney

Software-controlled With increased storage subsystem capacities and complexities, software-controlled


storage management is storage management is becoming a must. Software is quickly becoming an
becoming a must.
important enabler to keeping the explosive growth of data under control.
Figure 146 below outlines the leading storage software vendors’ market share and
growth rates.

144
The SAN Book III – October 22, 2001

Figure 146. Total Worldwide Storage Software Market Revenues by Selected Vendor
($ in millions)
1998 1999 2000
Market Market Market Change
1998 Share 1999 Share 2000 Share 1999-2000
EMC 444.7 15% 802.8 20% 1340.6 26% 67.0%
Veritas 293.2 10% 504.7 12% 855.1 16% 69.4%
IBM 607.7 21% 723.6 18% 843.9 16% 16.6%
Computer Associates 596.2 21% 798.7 20% 612.7 12% (23.3)%
BMC Software 164.9 6% 228.9 6% 224.1 4% (2.1)%
Network Appliance 14.9 1% 53.1 1% 156.0 3% 193.8%
StorageTek 86.9 3% 112.8 3% 143.5 3% 27.2%
Compaq 14.8 1% 44.9 1% 143.2 3% 218.9%
Legato 147.2 5% 172.2 4% 142.3 3% (17.4)%
Source: Dataquest (April, 2001)

Storage Software Product Descriptions by Category


We believe the best way to analyze the storage software market is within
Dataquest’s framework. We have been impressed with Dataquest’s analysis and,
therefore, use Dataquest’s definitions and estimates in most of our analysis.
However, we have added clustering as a separate category and have re-ordered the
categories in the way we prefer to look at the market. The following are detailed
descriptions of each enterprise storage software product category, in order from the
top of the software pyramid to the bottom (Figure 145 above), mostly provided by
Dataquest:
➤ Storage Resource Management: SRM products provide data collection and
automation agents that consolidate and operate on information from multiple
platforms supporting storage management tools on multiple operating systems,
and storage and SAN devices. Key functions include capacity reporting and
analysis, performance reporting and analysis, capacity and performance
management automation, storage management product integration, application
and database integration, and hardware integration. Basic network and system
management (NSM) integration should provide the ability of the SRM product
to externalize events to other management products via SNMP (storage network
management protocol). Product-specific integration includes the ability to
launch the SRM product from the NSM console. Integration with SAN and
device administration products and media management products should include
launch of hardware configuration utilities from the SRM console, collection and
reporting of agent information, and integration of logical-level data.
➤ Data Replication: This segment includes snapshot, data movement, and data
replication products.
➤ HSM and Archive: Hierarchical Storage Management (HSM) products operate
on defined storage policies that provide for the automatic migration of
infrequently used files to secondary storage. Archive products provide for the
storing of a point-in-time version of a file for historical reference.

145
The SAN Book III – October 22, 2001

➤ Backup/Recovery Network: This segment includes those products that


manage data protection processes on multiple systems from a single
management location.
➤ Backup/Recovery Host: This segment includes products designed specifically
to manage host-attached storage.
➤ Virtualization: Virtualization software creates an abstraction layer that
separates physical storage from logical storage, masks complexities to simplify
storage management and enables storage resource optimization via pooling.
This segment includes software products that sit on an application server or a
server appliance between the server and the storage.
➤ Clustering: Server clustering software creates highly available systems by
enabling groupings of servers to appear as one. If a primary server fails, a
secondary server will assume its functions.
➤ Media and Library Management: Products in this segment are designed to
manage and optimize drive and library hardware, removable media, and
operational labor across tape, optical platters, and CD-ROM. Media
management activities include allocating, labeling, tracking, recycling, and
monitoring media, as well as storage pool management.
➤ SAN and Device Administration: Storage subsystems and SAN infrastructure
component software products provide configuration utilities and agents that
collect capacity, performance, and status information, usually for a single device
type or a set of devices from a single vendor.
➤ Core Storage Management Technology: The core segment includes file
systems, volume management products, and emerging technologies that do not
fit into one of the other segments.

146
The SAN Book III – October 22, 2001

fs
Software Applications
➤ The four storage software applications we view as important
catalysts for future growth are: 1) Storage Resource Management, 2)
Replication, 3) Serverless and LANless Backup, and 4) Hierarchical
Storage Management. [We discuss file systems, volume
management, and virtualization in our previous section: “RAID Is No
Good Without Core Software.”]

➤ We expect to see these key four segments also drive growth in more
traditional storage software product areas.

➤ We believe EMC and Veritas are the two horses to bet on. Also, there
are other emerging companies coming out with new applications and
technologies that deserve attention.

The four storage software applications we view as important catalysts for future
growth are: 1) Storage Resource Management, 2) Replication, 3) Serverless and
LANless Backup, and 4) Hierarchical Storage Management. [We discuss file
systems, volume management, and virtualization in the previous “RAID Is No Good
Without Core Software” section.] We expect to see these key four segments also
drive growth in more traditional storage software product areas.
We believe EMC and Veritas are the two horses to bet on in the storage software
race. They are currently the storage software market leaders and both gained market
share in 2000. We also believe other companies are coming out with important new
products and technologies that merit attention.

1. Storage Resource Management (SRM)


SRM is one of the key We view SRM as one of the most coveted assets of emerging SAN environments.
areas of focus for SANs SRM is the looking glass into a data infrastructure that can enable greater
and represents excellent
growth opportunities.
manageability of resources. Therefore, it is high priced real estate (you see the logo
all day long) and it is very “sticky” (swapping applications is painful).
SRM improves manageability by allowing SAN administrators to view and control
products from multiple locations on a single console, versus managing them
individually. This can lead to significant cost savings by enabling four to seven
times greater data management per person, thereby, reducing the number of people
needed to manage the environment.
Additionally, the Graphical User Interface (GUI) is high priced real estate and is
very sticky. It’s similar to Windows. Once you get used to using Microsoft’s
Windows GUI, you might be more inclined to use Microsoft’s Word or Excel,
instead of (say) Word Perfect or Lotus 1-2-3. Controlling this real estate not only
opens up follow-on business, it also drives brand recognition (which is why the
vendor that built your computer monitor put its brand name on the front of the
screen).
This is ground zero of the storage battle and could prove to be a winner-takes-all
contest. Therefore, we believe SRM is very important to understand.

147
The SAN Book III – October 22, 2001

What Is SRM?
While SRM has traditionally been the software that provided a window to view the
storage environment from a single console and measure the performance of devices,
it is also increasingly including the device administration software that provides the
administrative tools to alter the environment, including policy definitions, device
zoning, load balancing, and capacity allocation. In other words, SRM used to be
defined as the software that made everything easy to view and measure. Now it
includes active management, referred to as “drag-and-drop” functionality. Think of
it in the context of your Microsoft Explorer application, separating viewing the
environment from using your mouse to make a change to it.
EMC’s ControlCenter 4.3 has been the clear leader in the Storage Resource
Management software market. We expect some new entrants to gain traction in
2002 and beyond, for example, Prisa’s VisualSAN, Veritas’s SANPoint Control 2.0,
InterSAN’s Virtual Private DataPath technology, Legato’s GEMS SRM, and Sun
Microsystems’ HighGround SRM 5.0.
Below in Figure 147 is a snapshot of EMC’s ControlCenter.
Figure 147. EMC’s ControlCenter Software

Third-party ESN Manager


Navisphere
switch software

Symmetrix Manager

Source: Salomon Smith Barney

Device Administration
Device Administration software controls a device; in other words, it can change its
environment — it is active, not passive. Brocade’s Fabric OS software is an
excellent example of device administration software. Another example is EMC’s
Symmetrix Manager, which is now tied together with ControlCenter.

148
The SAN Book III – October 22, 2001

Load Balancing
Brocade’s Fabric OS is Brocade’s Fabric OS is considered to be one of the more sophisticated storage
one of the more networking software device administration software applications. It helps create
sophisticated storage
networking software
network-aware storage environments. Features such as load balancing ensure the
device administration even flow of data through the network. This further contributes to data availability
software applications. by balancing traffic evenly across the network. In other words, if one switch (or
port) is being overloaded and another switch (or port) has excess capacity, some of
the data can be sent through the other, less congested switch (or port), resulting in a
more efficiently run network.
Trunking is also a component of load balancing. Trunking allows the aggregation of
the bandwidth from multiple ports. For example, four 2 Gpbs ports can be trunked
together to create a single virtual 8 Gbps port.

Failover
Another one of Brocade’s advanced device administration software features has
been its failover capabilities. This is the software that enables multiple networking
devices to reroute storage traffic through alternative paths upon the devices’ failure
(failover can also occur within the same device between multiple ports upon
component failure).

Security
Security has been increasing in importance to protect users from accessing data
without permission and protecting against corruptions and failures.
We believe Brocade’s advanced software features have been one of its strongest
assets that have helped it dominate the storage switch market.

Robust SRM Software Promotes Wider Adoption


We believe the availability of robust Storage Resource Management (SRM)
software is contributing to broader storage network adoption by providing a user-
friendly window to view and manage storage environments.
Figure 148 below is an example of an SRM GUI interface from Prisa Networks.
Note the similarity to Microsoft’s Explorer.

149
The SAN Book III – October 22, 2001

Figure 148. Prisa Networks’ VisualSAN Network Manager

Source: Prisa Networks

150
The SAN Book III – October 22, 2001

Figure 149 below has more examples of Prisa Networks’ GUI interface.
Figure 149. Prisa Networks’ VisualSAN Performance Manager

Source: Prisa Networks

SRM Projections and Market Share


Dataquest projects SRM software will grow to $1,709.5 million in 2005 from $271.3
million in 2000, representing a 44.5% CAGR (Figure 150 below).
Figure 150. Storage Resource Management Projections
Revenues in millions

$1,800
$1,600 44.5% CAGR
$1,400
$1,200
$1,000
$800
$600
$400
$200
$0
1998 1999 2000 2001 2002 2003 2004 2005
Source: Dataquest (May, 2001). CAGR is for 2000 through 2005.

151
The SAN Book III – October 22, 2001

While vendors such as Computer Associates, BMC Software, and IBM represent
mainframe SRM software and currently own the bulk of the market, the entrances of
Sterling (owned by Computer Associates), HighGround (owned by Sun), and EMC
(through its ControlCenter software) have pushed SRM into open systems and are a
main driver for Dataquest’s projections (see Figure 150).
Figure 151 below breaks out the market share of industry leaders on all platforms.
Figure 151. Storage Resource Management Market Share, All OS

1999 2000
HighGround (Sun) Others HighGround (Sun) Others
EMC
2% 17% 3% 14%
19% EMC
IBM IBM 33%
7% 7%

BMC Software BMC Software


13% Computer 18%
Computer
Associates Associates
42% 25%

Source: Dataquest (April, 2001)

Figure 152 below breaks out the market share of industry leaders on the UNIX
platform. EMC has the largest share of this market.
Figure 152. Storage Resource Management Market Share, UNIX

1999 2000
Hewlett-Packard Legato
2% Others
Others 1%
EMC 8%
18% HighGround (Sun)
31% 2%

Legato EMC
BMC Software
17% 54%
33%
Hewlett-Packard BMC Software
4% 27%
HighGround (Sun)
3%

Source: Dataquest (April, 2001)

Figure 153 below breaks out the market share of industry leaders on the Windows
NT platform. EMC gained significant share from 1999 to 2000.

152
The SAN Book III – October 22, 2001

Figure 153. Storage Resource Management Market Share, Windows NT

1999 2000
EMC Others
Others EMC
2% 23%
21% 31%

Northern Parklife Computer Northern Parklife


9% Associates 7%
48%
HighGround HighGround Computer
W. Quinn W. Quinn
7% 11% Associates
13% 13%
15%

Source: Dataquest (April, 2001)

EMC’s ControlCenter Is the SRM Leader


EMC’s ControlCenter is a suite of host-based software applications that provide user
management of storage components across EMC’s Enterprise Storage Networks
(ESNs). Management functions include monitoring, configuration, control, tuning,
and planning.
EMC’s ControlCenter has ControlCenter Characteristics
tight integration with the
➤ Tight Integration: ControlCenter works with the following subsystems:
hardware.
EMC’s Symmetrix and Clariion, Compaq’s StorageWorks, HDS’s Lightning,
and StorageTek’s Fibre Channel tape drives. ControlCenter also works with
Connectrix switches. Note: Connectrix switches are OEM’ed from McData and
Brocade. ESN (Enterprise Storage Network) Manager, a powerful plug-in
software application featuring zoning and path management, provides even
greater control over the switches. With regards to other vendors’ subsystems,
the viewing function should perform well but the management function could
have more difficulty tightly integrating due to the potential reluctance of
competitors to support EMC.
➤ Mainframe Platforms: In addition to working on open systems platforms,
EMC’s ControlCenter also works in mainframe environments, giving it access
to the entire market. EMC is one of the few storage software companies that has
a leading position on all three major platforms (mainframe, UNIX, and NT).

Veritas Has Entered SRM


Veritas has a long track Veritas’s SANPoint Control manages heterogeneous SANs to control devices —
record of successful
including volumes, HBAs, switches and storage subsystems — from a single
heterogeneous
solutions. management interface. Veritas had less than 1% of the SRM market in 2000 as
SANPoint was only released in the fourth quarter, but we expect it to expand
aggressively as it rolls out its 2.0 version released in May, 2001.
SANPoint Control Characteristics
➤ Heterogeneity: SANPoint Control 2.0 can specifically identify certain
subsystems from the same vendors as either EMC, HDS, or Compaq. It can also
detect certain switches from QLogic, Brocade, McData, and Inrange. Veritas
has a long track record of successful heterogeneous solutions since it has been
integrating its software with other vendors’ products for over a decade.

153
The SAN Book III – October 22, 2001

However, as a non-hardware vendor it does not have the advantage of being


tightly integrated with a proprietary subsystem.
➤ Open Platforms: SANPoint Control works on open systems platforms (UNIX
and NT).
➤ A Recent Entrant: Veritas only released SANPoint Control in the fourth
quarter of 2000. EMC released Symmetrix Manager, an early incarnation of
ControlCenter, in early 1998. A more mature product can offer greater
reliability through years of testing in actual environments.

2. Replication
Replication is We view replication as a critical business asset. Replication creates multiple copies
transitioning from of data to ensure an uninterrupted business environment. A large part of replication
becoming a
discretionary, to a
is that it acts as an emergency plan for data infrastructures that can prevent a
necessary, expense. temporary or permanent loss in data availability. Therefore, replication is
transitioning from becoming a discretionary, to a necessary expense and in some
cases is being leveraged as a competitive advantage as data grows both in quantity
and significance and customers demand access to information 24 hours a day, seven
days a week without interruption for any reason.
Replication has been a solid area of growth for EMC in the past and we think it’s
only going to get better.
What Is Replication?
Replication enables Replication enables the mirroring of data between disk storage subsystems to ensure
mirroring of data the continuous availability of enterprise data during planned or unplanned outages.
between two physically
and geographically
Replication provides business continuance capability in the event of a data center
separated storage disaster, and during planned events such as daily backups, database loads and
subsystems. refreshes, application testing, scheduled maintenance, and data center migrations or
consolidations.
Below we have illustrated what data replication might look like.
Figure 154. Data Replication

Primary Data

Replicated Copy

Source: Salomon Smith Barney

154
The SAN Book III – October 22, 2001

Although Veritas recently introduced its own replication solution, EMC is still the
clear leader in this market segment. Other players include Legato’s Octopus,
NetApp’s SnapMirror, and NSI’s Double-Take.
Disaster Recovery (DR)
Disaster Recovery software, a subset of Replication, enables the high availability
(HA) of data by creating a real-time copy at a remote location. In the event of a
disaster, such as a power outage, which could affect the data infrastructure in a
particular business or region, businesses can be switched over to a replicated data
facility in a different location.
Without replication, lost data and prolonged downtime could result in a loss of
massive amounts of revenue and productivity, as well as in customer trust, brand
equity, and competitive advantages which can take years to build but just hours to
destroy. Replication can also prevent a loss in productivity as employees sit idle for
hours while the original data infrastructure is restored, as would be the case with a
recovery from a tape backup.
Below in Figure 155, we illustrate how a disaster recovery architecture might be
viewed.
Figure 155. Disaster Recovery Architecture

Primary data
center in NY

Backup data
center in FL

Source: Salomon Smith Barney

Figure 156 below provides estimates of the cost of downtime.

155
The SAN Book III – October 22, 2001

Figure 156. Downtime Costs

Application Downtime Costs


Financial Brokerage $6,450,000 / hr
Credit Card Authorization $2,600,000 / hr
Home Shopping $113,000 / hr
Catalog Sales $90,000 / hr
Airline Reservations $90,000 / hr
Tele-Ticket Sales $69,000 / hr
Package Shipping $28,000 / hr
Source: Fibre Channel Industry Association

Replication Projections and Market Share


We expect replication to be one of the fastest growing storage software segments.
Dataquest projects replication software will grow to $3,323.2 million in 2005 from
$966.9 million in 2000, representing a 28.0% CAGR (see Figure 157 below). As
bandwidth and disk prices decline, the investment needed to create a secondary site
for disaster recovery decreases.
Figure 157. Data Replication Projections
Revenue in millions

$3,500
$3,000 28.0% CAGR

$2,500
$2,000
$1,500
$1,000
$500
$0
1998 1999 2000 2001 2002 2003 2004 2005
Source: Dataquest (May, 2001). CAGR is for 2000 through 2005.

Data Replication has been one of EMC’s most valuable assets, generating 42% of
EMC’s new software license revenues in 2000. According to Dataquest, EMC’s
replication software has dominated its segment with 58% market share in 2000,
while its three closest competitors combined represented only 19% (see Figure 158
below). According to Dataquest, EMC grew its replication revenues 38% in 2000.
Note: EMC replication revenues include SRDF, TimeFinder, InfoMover,
CopyPoint, and Extractor/Resource Pak.

156
The SAN Book III – October 22, 2001

Figure 158. Replication Market Share, All OS

1999 2000

Others Others
StorageTek
19% 19%
4%
StorageTek
5% Symantec
4%

Symantec Compaq
6% EMC
EMC 5% 58%
Compaq 65%
Network
1% Appliance
Network 10%
Appliance
4%

Source: Dataquest (April, 2001)

Figure 159 below illustrates the UNIX replication market share. EMC clearly
dominates this market segment.
Figure 159. Replication Market Share, UNIX

1999 2000
Others Others
11% 15%

EMC
EMC
85%
89%

Source: Dataquest (April, 2001)

Figure 160 illustrates the Windows NT replication market share. EMC has
significantly increased its lead in this segment.

157
The SAN Book III – October 22, 2001

Figure 160. Replication Market Share, Windows NT

1999 2000
Others Others
21% EMC 15%
29% Legato EMC
9% 41%

PowerQuest
Legato
11%
21% Compaq
PowerQuest 13% Compaq
16% 24%

Source: Dataquest (April, 2001)

EMC’s SRDF Is the Clear Leader


Take a Look at EMC’s Secret Sauce — SRDF
EMC’s replication EMC’s replication software, Symmetrix Remote Data Facility (SRDF), has been one
software, SRDF, has of its key differentiators. SRDF is EMC’s remote data replication software that
been one of its key
differentiators.
enables the synchronous replication of data in various locations. In our opinion,
ensuring disaster recovery is one of the single most important assurances an IT
manager wants.

SRDF Characteristics
➤ Storage-centric: SRDF provides this facility simultaneously to mainframe,
UNIX, Windows NT, and other systems. This storage-centric (i.e., residing on
the subsystem), mirrored, data storage solution duplicates production site data
on one or more physically separate target Symmetrix systems. Systems can be
across the room, across the globe, or anywhere in between.
➤ Low Transmission Costs: SRDF’s multi-hop capability decreases line
transmission costs by mirroring only the tracks changed since the last update,
and by taking advantage of lower cost transmission lines, such as Ethernet.
➤ Simultaneous Use of Data: Scheduled backup operations, data warehouse
loading, or new application testing can be performed from the target Symmetrix
data (the second copy), while normal operations are continuously running on the
original Symmetrix system.
➤ Synchronous and Asynchronous: EMC mainly offers SRDF over Fibre
Channel, interconnected by T1 and T3 lines, for synchronous applications.
EMC also offers SRDF over IP (Internet Protocol) for asynchronous
applications as a less expensive, less robust solution. (Note: Synchronous is real
time, asynchronous is for static information.)
Below is an illustration of a potential architecture using EMC’s SRDF. In one
branch, two Symmetrix systems are connected via dark fiber. In this configuration,
replication can be accomplished either synchronously or asynchronously. In another
branch, two Symmetrixes are attached via IP. In this configuration, replication can
only be accomplished asynchronously.

158
The SAN Book III – October 22, 2001

Figure 161. EMC’s Replication Software Can Mirror Data in Two Different Locations

Tape
Library

EMC’s SRDF over


Dark Fiber
EMC
Client 1 Application Sym 1 EMC
Server 1 Symmetrix 4

EMC
Client 2
LAN SAN Symmetrix 2

EMC’s SRDF over IP


Application EMC
Client 3 Server 1 Symmetrix 5
EMC
Symmetrix 3 WAN/ MAN/
LAN/ SAN
EMC EMC’s EMC
Celerra SRDF Symmetrix 6
over IP

Source: Salomon Smith Barney

Veritas Has Entered Replication


We believe Veritas is in an excellent position to benefit from replication market
expansion over the next few years. Why? Aside from Veritas’s increasing focus on
this market segment, it can provide heterogeneous replication across platforms,
servers, and storage (EMC only provides the first two and, we would argue, does not
currently have much incentive to help out its competitors by providing the third).
We believe this creates significant opportunity for Veritas.

Veritas Is Poised to Rock Replication


We believe vendors are In our opinion, 2002 will be the year Veritas really gains replication momentum.
looking for a third-party Note: In addition to its primary Volume Replicator product, Veritas also offers two
vendor for replication
software. We believe
other replication products: File Replicator and Storage Replicator for Windows.
Veritas fits that bill.
Volume Replicator Differences
➤ A Recent Entrant: Veritas released Volume Replicator in the third quarter of
2000 (although the company had been developing the product for many years
before it was released).
➤ Server-centric: Volume Replicator is server-centric (i.e., it resides on the
server) as opposed to EMC’s SRDF, which resides in the storage subsystem.
➤ Heterogeneity: Volume Replicator works with storage subsystems from
multiple vendors, offering greater vendor flexibility. EMC’s SRDF replicates
data from one Symmetrix (its subsystem offering) to another Symmetrix. A
homogeneous solution can offer greater functionality since it is more tightly
integrated with the unique features of a particular hardware platform; however,
customers often prefer flexibility in choosing subsystems from multiple vendors.

159
The SAN Book III – October 22, 2001

3. Serverless and LANless Backup


We believe the real We believe that serverless and LANless backup are two of the most compelling
“killer application” that reasons for SAN adoption. Serverless and LANless backup are high availability
SANs can deliver today
is serverless and
methods to back up data. By leveraging the SAN architecture, these applications
LANless backup. free up expensive bandwidth and CPU cycles. Therefore, serverless and LANless
backup are “killer applications,” which reduce infrastructure costs and maximize
available resources by offloading work from the LAN to the SAN.
Serverless and LANless backup improve availability by allowing the backup of data
without the need for a dedicated server or without having to utilize the LAN. This
can lead to significant cost savings by requiring less servers and bandwidth for the
data infrastructure.
We believe these applications will be two of the largest drivers for SAN adoption.
This could prove to be a high growth area for storage software companies.

Traditional DAS Backup


In a traditional DAS architecture, generally speaking, application servers and the
LAN must be used when backing up data. Large data backups can sap an
application server and congest the LAN. Therefore, by offloading backups onto a
SAN, servers and LANs can function more efficiently without the burden of bulk
storage traffic.
In Figure 162 below, backing up Subsystem 2 to the tape library traditionally would
send the data through Server 2, onto the LAN, through Server 1, and then into the
tape library (the reverse would occur to recover the data).
Note: The thick lines represent the data path in the figures that follow.

160
The SAN Book III – October 22, 2001

Figure 162. Traditional DAS (Direct Attached Storage) Backup

Tape
Library
DAS
(Traditional Storage Backup)
Client 1
Server 1 •In a traditional DAS
architecture, backup traffic
Subsystem 1
must traverse an already
Client 2
LAN congested LAN (often at
night or during non-business
Server 2
hours).
Client 3 Subsystem 2
•DAS backups involve
application servers to run the
backup application software.
Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

What Is LANless Backup?


SANs enable LANless SANs enable LANless backup by offloading backup traffic onto a dedicated storage
backup by offloading network thereby reducing LAN congestion (see Figure 163).
backup traffic onto a
dedicated storage Figure 163. SAN Backups Can Be LANless
network thereby
Tape
reducing LAN Library
congestion.
SAN
Client 1
Server 1
(LANless Backup)
•In a SAN architecture,
Client 2
Subsystem 1 backup traffic can be
LAN SAN offloaded onto a dedicated
storage network which
Server 2 avoids further LAN
Client 3 congestion.
Subsystem 2

•LANless backup does


involve an application server
Client 4
Server 3
to run the backup software.
Subsystem 3

Source: Salomon Smith Barney

Many traditional architectures avoid this type of LAN congestion by directly


connecting servers to each other via a SCSI or Fibre Channel cable. We would
argue that this is an early stage SAN. The next logical evolutionary step is
serverless backup as intelligence is added into the network and onto subsystems.
What Is Serverless Backup?
Taking backup one step further, SANs enable serverless backup by offloading
backup traffic onto a dedicated storage network to reduce LAN congestion as well as

161
The SAN Book III – October 22, 2001

server involvement (see Figure 164). Therefore, serverless backups are also
LANless by definition.
Figure 164. SAN Backups Can Be LANless and Serverless

Tape
Library
SAN
Client 1 (Serverless Backup)
Server 1
•In a SAN architecture,
backup traffic can be
Subsystem 1 offloaded onto a dedicated
Client 2
LAN SAN storage network to
avoid further LAN
Server 2 congestion.
Client 3
Subsystem 2
•Backup and recovery do
not have to involve an
application server to run the
Client 4
Server 3 backup application software,
Subsystem 3 thereby freeing server
capacity.

Source: Salomon Smith Barney

We believe Veritas’s NetBackup, Legato’s Celestra, and Computer Associates’


ARCserveIT are the leading LANless and serverless backup software products. We
expect the availability of these killer applications (particularly serverless backup)
will significantly increase the demand to network storage. Although these
applications have been promised for over a year, we expect broad adoption to begin
in 2002, as other SAN technologies also evolve.

4. Hierarchical Storage Management (HSM)


Hierarchical Storage We view Hierarchical Storage Management (HSM) as one of the traditional
Management (HSM) is applications which can benefit greatly from increased SAN adoption. The goal of
one of the traditional
applications which can
HSM is to reduce the cost of storing data by more effectively utilizing secondary
benefit greatly from storage capacity. Secondary storage can cost as little as 3% of the cost of primary
increased SAN adoption. storage.
HSM can leverage storage capacity by more efficiently incorporating secondary
storage; thereby potentially expanding a user’s available storage capacity with
existing resources (i.e., without increasing cost). Secondary storage capacity,
particularly tape, is relatively less expensive (and easier) to expand. Don’t you hate
getting reminders to empty out your email basket and then being locked out of
sending new emails because you have exceeded your storage limit? Then, after
deleting all old emails, you discover that you need data from an old email. Read on
to see how HSM could help.
HSM Positives:
➤ It reduces costs
➤ It expands storage capacity

162
The SAN Book III – October 22, 2001

HSM Negatives:
➤ Slower data access
SANs enable a broader application of HSM, even though HSM has been around for
a while. SANs make HSM more attractive by reducing the resources needed to
utilize it.
HSM is a good example of how the adoption of a SAN infrastructure can make
traditional applications even more powerful. Additionally, as IT budgets come
under more scrutiny, applications with immediate and tangible cost savings should
grow in importance. Therefore, we believe HSM is important to understand.
What Is HSM?
Hierarchical Storage Hierarchical Storage Management is just what it sounds like: prioritized data
Management (HSM) is storage. HSM assigns a level of “usage” to data. By doing so, intelligent HSM
just what it sounds like:
prioritized data storage.
software can prioritize the accessibility of data and store it according to
predetermined parameters. For example, HSM would locate a frequently used file
(say a phone list) where it could be readily accessed on a primary storage device.
On the other hand, HSM might migrate less frequently used data (with
predetermined and preset parameters and thresholds; for example, an employment
contract) to a secondary storage device. In general, HSM is designed to migrate less
frequently used data from primary, quick-access storage to secondary, less
expensive, slower-access storage. Below are definitions of primary and secondary
storage:
Primary storage
A primary storage device is a storage device or subsystem that provides fast,
easy access to data, such as a disk subsystem.
Secondary storage
A secondary storage device is a storage device or subsystem that provides
less expensive, slower access to data, such as a tape library or optical
jukebox.
In Figure 165 below, we have outlined how HSM might view data.

163
The SAN Book III – October 22, 2001

Figure 165. An Example of How HSM Views Data

Disk Subsystem Disk Subsystem

Unused Capacity Unused Capacity

The way
HSM Less Frequently
Looks At It Used Data
Data
Frequently
Used Data
Source: Salomon Smith Barney

The goal is to reduce the HSM can migrate data based on many different variables, or by combining
cost of storing data by numerous variables, such as frequency of use, date (i.e., file age), file size, and type
more effectively utilizing of data (e.g., attachments, graphics). An example might be email attachments and
secondary storage
capacity. Secondary
emails two weeks old and older. In this example, all attachments and emails that are
storage can cost as little two weeks old or older would be migrated from a disk storage subsystem to a tape
as 3% of the cost of library. The goal is to reduce the cost of storing data by more effectively utilizing
primary storage.
secondary storage capacity.
OTG Software provides a The cost advantage of HSM does not come without a performance penalty. Here
unique indexing scheme again, we run into the importance of data access, rather than data storage. This
that truncates (migrates)
functionality can be a key differentiator. For example, OTG Software provides a
target data from its
location on primary unique indexing scheme which truncates (migrates) target data from its location on
storage, leaving behind a primary storage, leaving behind a small “stub file.” The stub file has the updated
small “stub file.” location of the target data. When clients access data that has been migrated, they hit
the stub file, which, in turn, accesses the target data directly from secondary storage
without first having to copy it back to primary storage. In Figure 166 below, we
have illustrated how HSM effectively maximizes storage capacity.

164
The SAN Book III – October 22, 2001

Figure 166. HSM Data Migration — Freeing Primary Storage Capacity by Utilizing Less Expensive
Storage

Tape Library
Disk Subsystem Disk Subsystem

Unutilized Tape Backup


Unused Capacity

Using HSM
Unused Capacity
Data
Less Frequently Migration
Used Data Stub Files HSM Migrated Data

Frequently Frequently Utilized Tape Backup


Used Data Used Data

Source: Salomon Smith Barney

SANs Make HSM a “Killer App”


We believe the deployment of SANs, which provide an any-to-any architecture,
increases the attractiveness of HSM by enabling more seamless primary-to-
secondary storage data migrations, without involving the LAN and application
servers.
By taking advantage of a By taking advantage of a SAN architecture, HSM becomes much more seamless
SAN architecture, HSM and, therefore, a “killer application.” In a DAS architecture, data must traverse
becomes much more
seamless and, therefore,
servers and LANs in order to gain access to a tape library (i.e., a typical HSM data
a “killer application.” migration). This may be too taxing on servers and LANs to gain any benefit. In a
SAN architecture, data migrations from disk to tape can be done more efficiently by
utilizing a dedicated storage network.
Below in Figure 167 we illustrate data migration from a disk storage subsystem
(primary storage) to a tape library (secondary storage) in a traditional Direct
Attached Storage (DAS) architecture.

165
The SAN Book III – October 22, 2001

Figure 167. Data Migration in a DAS Architecture

Tape
Library
DAS
(Traditional Storage Backup)
Client 1
Server 1 •In a traditional DAS
architecture, backup traffic
Subsystem 1
must traverse an already
Client 2
LAN congested LAN (often at
night or during non-business
Server 2
hours).
Client 3 Subsystem 2
•DAS backups involve
application servers to run the
backup application software.
Client 4
Server 3
Subsystem 3

Source: Salomon Smith Barney

Below, in Figure 168 we illustrate the same data migration as above, but with a SAN
architecture.
Figure 168. Data Migration in a SAN Architecture

Tape
Library
SAN
Client 1 (Serverless Backup)
Server 1 •In a SAN architecture,
backup traffic can be
Subsystem 1 offloaded onto a dedicated
Client 2
LAN SAN storage network which
avoids further LAN
Server 2 congestion.
Client 3
Subsystem 2
•Backup and recovery does
not have to involve an
application server to run the
Client 4
Server 3 backup application software,
Subsystem 3 thereby freeing server
capacity.

Source: Salomon Smith Barney

Using a SAN, no servers or LAN need to be accessed for tape backup, increasing the
availability of these resources for other uses.

166
The SAN Book III – October 22, 2001

HSM Projections and Market Share


Dataquest projects HSM and Archive software will grow to $603.8 million in 2005
from $307.1 million in 2000, representing a 15% CAGR (Figure 169 below). Note:
Archive products create a point-in-time version of a file which can be retrieved at a
later date.
Figure 169. HSM and Archive Projections
Revenue in millions

$700
$600 15% CAGR
$500
$400
$300
$200
$100
$0
1998 1999 2000 2001 2002 2003 2004 2005
Source: Dataquest (May, 2001). CAGR is for 2000 through 2005.

Figure 170 below shows HSM and Archive market share. IBM dominates this
market segment.
Figure 170. HSM and Archive Market Share, All OS

1999 2000
Others Others
25% 27%

IBM IBM
49% 48%

SGI SGI
12% 8%

Computer OTG Software Computer OTG Software


Associates 4% Associates 9%
10% 8%

Source: Dataquest (April, 2001)

Figure 171 below shows HSM and Archive market share on the UNIX platform.
The market is fairly fragmented with no one having more than 16% market share.

167
The SAN Book III – October 22, 2001

Figure 171. HSM and Archive Market Share, UNIX

1999 2000
FileTek
FileTek
15%
17% Others
Others
36% 37% SGI
SGI
14%
12%

Veritas Legato
IBM IBM
3% 3%
11% Sony ADIC 12%
Legato ADIC
Veritas 5% 9%
Sony 8% 10%
5%
3%

Source: Dataquest (April, 2001)

Figure 172 below illustrates HSM and Archive market share on the Windows NT
platform. Computer Associates and OTG Software are the leaders in this market.
Figure 172. HSM and Archive Market Share, Windows NT

1999 2000
Others
Others OTG Software 20%
22% 25%
KOM Networks OTG Software
4% 42%

Smart Storage
KOM Networks
13%
3%
Computer Computer
Smart Storage
Associates Associates
11%
39% 21%

Source: Dataquest (April, 2001)

OTG Software’s HSM Products


We believe OTG is the We believe OTG is the leader in HSM software, with Veritas its main competitor.
clear leader in HSM As we see it, the benefits of OTG’s technology are that it is more leveragable and
software, with Veritas its
main competitor.
flexible than competitors’ offerings. We believe OTG has achieved its leadership
position due to its intent focus on the HSM market.
We believe OTG’s EmailXtender is a “killer application” that will provide leverage
for its other Xtender products. OTG’s EmailXtender can uniquely provide: 1)
HSM, 2) the ability to easily perform key word and full text searches, and 3) the
ability to choose multiple secondary storage mediums, such as optical. From what
we have seen, competing products cannot perform all these tasks. OTG’s
EmailXtender can reduce file or data searches from about 11 hours, or from about
five hours if the target data has been electronically backed up, to seconds. We
believe this bodes well for OTG in today’s increasingly regulated environment
which mandates that data be stored more often and for longer periods.
OTG has also developed industry-leading content management and search/archive
functionality into EmailXtender. We believe that these core features, combined with
its traditional HSM technology, are what truly puts OTG Software above others.

168
The SAN Book III – October 22, 2001

Veritas’s HSM Solution


While Veritas only captured 1.5% of the HSM market in 2000, according to
Dataquest, we believe it is (and will continue to be) OTG’s most visible competitor.
Further, we believe Veritas is going to continue to aggressively target the HSM
market over the upcoming years.
Based on technology, we believe OTG Software has a superior solution. As outlined
in the sections above, OTG’s technology migrates data based on a predetermined set
of attributes. In other words, OTG’s technology is flexible enough to migrate old
data, large files, attachments, or data measured by various other predetermined
attributes. Veritas’s HSM software is not as flexible. For example, Veritas’s email
product (Remote Storage for Exchange or RSE) only manages email attachments. In
our opinion, OTG has the best HSM technology in the market.
Veritas has a broader distribution outlet and the most popular backup software,
which could make it a formidable competitor. Although, on a product basis, OTG
seems to have the edge, Veritas has a broader market footprint spanning a more
comprehensive product portfolio. Veritas sells its HSM product packaged (or
integrated) with many other products, including its popular Netbackup software.
OTG sells its products as an add-on. This could put Veritas at a competitive
marketing advantage. However, OTG’s partners, Compaq, Legato, Maxtor,
StorageTek, Tivoli/IBM, and Quantum, are all OEMs of OTG and embed its HSM
technology in one form or another, which could level the playing field.
On July 16, 2001, Veritas announced it had created an HSM solution to manage
Microsoft Exchange e-mail attachments for the Maxtor MaxAttach NAS 4100.

169
The SAN Book III – October 22, 2001

This page intentionally left blank.

170
The SAN Book III – October 22, 2001

Networking

Storage Networking
Storage Networking I/O
Storage Networking Equipment

171
The SAN Book III – October 22, 2001

Storage Networking
➤ We believe storage networking is on the cusp of resuming solid
growth in 2002, due to many product launches, such as Brocade’s
Silkworm 12000, and the industry’s migration to 2 Gbps and PCI-X.

➤ We expect storage networking companies to lead the storage


industry’s growth.

➤ In this section we provide a market overview, followed by more


detailed discussions on HBA and storage networking technologies,
markets, and companies.

Peeling the Onion — It’s Not Really a Cloud


In reality, the “SAN Q: Why is the SAN pictured as a cloud?
cloud” is composed of
many different types of A: Because the real depiction (illustrated below) hurts to look at.
networking equipment.
In reality, the “SAN cloud” is composed of many different types of networking
equipment; including HBAs, hubs, routers, switches, and Directors.
Figure 173. The Secret Behind a SAN Cloud

Disk 1
Server 1
Disk 2

Server 2
Tape 1

Server 3 Disk 3
LAN

Disk 4
Server 4
Disk 5

Server 5 Disk 6

Server 6 Disk 7

Disk 6

NAS
Server 1 Disk 7

NAS
Server 2 Tape 2

Disk 8

Source: Salomon Smith Barney

Note: In the diagram we use switches, which are usually purchased with 16 ports
(moving to include 32 ports). We could use Directors instead or as well. Directors

172
The SAN Book III – October 22, 2001

come with 64 ports (McData) and 128 ports (Inrange). We do not specifically depict
any routers or hubs per se. Also, note the redundancies in order to ensure multiple
paths and greater uptime in case of a path or device failure.

Mommy, Where Do SANs Come from?


Once upon a time there were mainframes and they ruled the earth. Mainframes were
(and still are) large scale supercomputers. Think of them as huge servers – IBM
does. A challenge of the mainframe was storing and retrieving large amounts of
data. Here they faced two issues. First was that the IBM mainframe S/390, for
example, only has 256 maximum ESCON channels for storage connectivity.
However, mainframes (for the most part) can have greater storage requirements than
their 256 ports can supply. The second issue was that multiple mainframes needed
access to common disk and tape storage subsystems – remember, these are
expensive items.
Therefore, IBM introduced Enterprise System CONectivity (ESCON), a new and
faster system interconnect (at the time) that could attach to multiple systems. Note:
ESCON is simply mainframe connectivity, much the same way SCSI is for open
systems (UNIX and NT). To manage this storage traffic, ESCON Directors were
created by IBM, Inrange and McData to enable greater mainframe connectivity.
ESCON Directors showed up on the market back in the early 1990s. So, storage
networking is not all that new of a concept. Notice that the capital “D” in Directors
and the name itself is derived from the ESCON Director (mainframe) market.

Today’s SANs Are Simply an Extension of Yesterday’s Mainframe


Storage Architectures Combined with Data Networking Expertise
For mainframes, ESCON Directors were placed behind the mainframe in order to
“fan out” and enable greater storage connectivity; thereby, increasing performance
and flexibility. In a sense, today’s SANs are simply an extension of yesterday’s
mainframe storage architectures, applied to a wider variety of computers and
combined with data networking expertise. Note: Some would argue it’s more of a
networking model applied to storage.
Inrange and McData have come from the mainframe world and are therefore,
focused more on providing higher-end storage networking (i.e., Directors). Brocade
and QLogic (through its Ancor acquisition) have come to the market with more of a
data networking background and are therefore focused on the mid-range and low-
end markets (i.e., switches).
Below we have illustrated a mainframe storage architecture.

173
The SAN Book III – October 22, 2001

Figure 174. Mainframe Storage Architecture Using ESCON Directors

Storage
Mainframe ESCON
Directors

Source: Salomon Smith Barney

Market Breakdown
SAN is one of the most We believe the SAN concept is one of the more exciting killer applications emerging
intriguing killer in today’s information-centric world. But what actually goes on inside the SAN
applications to emerge.
cloud?
We have positioned the various storage networking components in Figure 175
below, with the upper portion of the pyramid representing more
sophisticated/higher-cost devices. We have also depicted the segments we believe
are current growth areas. Figure 175 essentially represents our view of the storage
networking market.

174
The SAN Book III – October 22, 2001

Figure 175. SAN Networking Component Analysis

SAN
APPLIANCE

DIRECTOR

Technology
SWITCH

Cost

R
TE
(FABRIC)

U
RO
SWITCH (FC-AL)
BRIDGE
HUB

HBA

Current Growth Categories Connected Product Categories OEM Products

Source: Salomon Smith Barney

Note: The terms and names of storage networking equipment are very similar to
LAN and WAN networking equipment (i.e., bridges, routers, hubs, and switches) but
are, in fact, very different technologies.

Projections
Storage Networking Growth
Storage networking has emerged as one of the highest-growth areas of storage. We
expect storage networking growth to lead the overall storage industry. According to
Dataquest, storage networking growth (2000–05) is expected to reach a 67.0%
CAGR versus storage software’s estimated 26.1% CAGR and RAID-based disk
storage 14.7%.

175
The SAN Book III – October 22, 2001

Figure 176. Storage Networking Growth Is Expected to Lead the Industry (YoY Growth Projections)

250%

200%

150%

100%

50%

0%
1999 2000 2001 2002 2003

Networking Software RAID

Source: Dataquest (May, 2001)

Projections
Dataquest projects storage networking revenues will grow to $16.9 billion in 2005
from $1.3 billion in 2000, representing a 67.0% CAGR. This projected high-growth
market has attracted fierce competition from many companies looking to be the
Cisco of storage — including Cisco. (See the “Fibre Channel Versus Ethernet”
section for more details).
Figure 177. Projected Storage Networking Revenue Growth

CAGR
(Revenue in $millions) 2000 2001 2002 2003 2004 2005 2000-2005
FC Routers $50 $70 $110 $160 $240 $360 48.4%
FC Hubs $50 $60 $60 $60 $60 $60 3.7%
FC Switches $470 $710 $1,090 $2,210 $4,090 $4,930 60.0%
FC Integrated Switches $0 $90 $180 $460 $840 $1,040 NM
FC HA Core Products $190 $400 $1,040 $2,390 $4,620 $6,240 101.0%
FC HBAs $540 $810 $1,140 $2,090 $3,480 $4,250 51.1%
Total Storage Forecast $1,300 $2,140 $3,620 $7,370 $13,330 $16,880 67.0%
Year-to-Year Growth -- 64.6% 69.2% 103.6% 80.9% 26.6% --
Source: Dataquest (May, 2001)

Platform Focus
Windows NT is clearly expected to be a SAN growth driver. This should bode well
for Brocade (which we believe has the lion’s share of the NT storage networking
market), Emulex (which supplies Fibre Channel HBAs to Compaq, and EMC), and
QLogic (which supplies Fibre Channel HBAs to Dell and NetApp).

176
The SAN Book III – October 22, 2001

Using Dataquest’s JOBD and external RAID Fibre Channel storage estimates and
projections as a reference point, we estimate that NT should grow from
approximately 30% of the SAN market in 2000 to approximately 57% in 2005.
Note: Although UNIX is projected to surrender market share to NT, what is not
captured below is the fact that NT is also growing nicely. In other words, NT is
simply expected to outpace UNIX growth. Also, since FICON has only recently been
released, it has not yet been included in this analysis.
Figure 178. SAN Market Share by Platform

2000 2005

Other
Other 1%
3%
NT
30%
UNIX
42%

NT
57%

UNIX
67%

Source: Dataquest and Salomon Smith Barney

177
The SAN Book III – October 22, 2001

Storage Networking I/O: HBAs and ICs


➤ An HBA is needed with most external server-to-storage connections.
Therefore, HBA growth should ride the wave of migrations from
internal to external storage. This — combined with more HBA
business moving to independent (third-party) players, increasing
Fibre Channel adoption, and the emergence of additional protocols
(such as iSCSI and Infiniband) — should propel the HBA vendors
with solid growth.

➤ Although HBAs are also used in non-networked storage


environments, SANs increase their importance and demand greater
functionality.

The Fibre Channel HBA Market


The Fibre Channel Host The Fibre Channel Host Bus Adapter industry has experienced phenomenal growth
Bus Adapter industry in recent years, reaching $540 million in 2000, according to Dataquest. The five-
has experienced
phenomenal growth in
year CAGR through 2005 is expected to reach 51.1% (notice that much of this
recent years. growth is front-end loaded).
Figure 179. Fibre Channel Host Bus Adapter Growth Projections
Revenues in millions

$4,500
$4,000 51.1% CAGR
$3,500
$3,000
$2,500
$2,000
$1,500
$1,000
$500
$0
2000 2001 2002 2003 2004 2005

Source: Dataquest

178
The SAN Book III – October 22, 2001

HBA Growth Drivers


The growth drivers of the HBA industry have provided an excellent backdrop for all
members of the group (i.e., a rising tide has lifted all boats), each of whom has
focused on different segments of the market.
Growth drivers include:
1 the migration to external storage (which uses HBAs, as opposed to internal
storage, which does not),
2 the industry is quickly moving to independent (or non-captive or third-party)
players,
3 the greater proliferation of Fibre Channel networking,
4 the increasing demand for greater HBA functionality,
5 the emergence of additional protocols (such as iSCSI and Infiniband), which we
expect to be at least partially serviced by the leading Fibre Channel HBA vendors
(all of which have initiatives underway to develop new protocol interface cards
for future demand).

What Is an HBA?
A Host Bus Adapter Much like a Network Interface Card (NIC) interconnects servers to PCs via the LAN
(HBA) interconnects or Ethernet, a Host Bus Adapter (HBA) interconnects servers to external storage
servers-to-external
storage devices.
devices (note: HBAs provide far more functionality than NICs, which we discuss in
the “Anatomy of HBAs” section). Back when storage was internal to the server, no
such Input/Output (I/O) interface was required, since storage was not externally
attached. With the advent of external storage, I/O interfaces (i.e., HBAs) became
necessary to provide inter-device communication.
Below in Figure 180, we have outlined the placement of NICs and HBAs within
server-to-PC and server-to-storage architectures. Note: since internal storage is
inside the server, there is no such need for an I/O interface.
Figure 180. HBAs Emerged as a Server-to-Storage I/O Interconnect

LAN Server
SAN Storage
Subsystem

PC

Fibre Channel
Ethernet

NIC HBA

Source: Salomon Smith Barney

179
The SAN Book III – October 22, 2001

Below we have listed a side-by-side comparison between NICs and HBAs.


Figure 181. HBAs Versus NICs

Handled Handled
Function by HBA by NIC

Framing packets ✔ ✔
Provide MAC address ✔ ✔
Link level error checking ✔ ✔
Sequence level error checking ✔
Manages flow control ✔
Performs SAR ✔ Requires
Server
Provides linking of I/O request and packet ✔ CPU
Handles chaining of multiple requests ✔
Manages many concurrent I/O ✔
Ability to analyze and manage data ✔
Storage virtualization ✔

Source: Emulex

Digging a Bit Deeper into Storage ICs


More often than not, I/O In physical terms, the link between the system and the external storage device needs
refers to storage. a controller on both the host side (within a server) as well as the target side (within a
storage peripheral or subsystem). The two controllers negotiate the transmission of
data and must use the same protocol (i.e., IDE, SCSI, or Fibre Channel).
On the host side (the server), the interface can appear as an HBA or an integrated
circuit (IC) on a motherboard. In the Ethernet world, Intel and other chipset vendors
standardized IDE and its variants (EIDE, UDMA) into their chipsets; most PC
systems today already include the IDE interface technology embedded. For higher
performance computing, however, SCSI often appears in systems as either an HBA
or as a separate chip on a motherboard. Fibre Channel, due to its high performance,
also often (although not exclusively) appears in the form of an HBA.
On the peripheral or target side (the storage), the interface typically appears as an
independent controller IC or embedded within a single chip solution and hardwired
into the storage subsystem on its disk drives (the exception is IBM’s Shark storage
subsystem that uses an Emulex HBA for FICON). For example, the hard disk drive
controller ASIC produced by ST Micro or Lucent embeds either a SCSI or IDE
interface controller. The ASIC is then used in Seagate and Quantum disk drives. In
contrast, both Fujitsu and Hitachi currently utilize the QLogic target interface IC
within their Fibre Channel disk drives. The trade-offs of embedding a solution
versus using off-the-shelf silicon are clear; off-the-shelf components allow OEMs to
get to market faster, but add to cost.

180
The SAN Book III – October 22, 2001

Below in Figure 182 we show the relationship between the host controller and the
target controller.
Figure 182. Host and Target Controllers Negotiate Data Transfer

Server Storage
Subsystem

Disk Drives

Fibre Channel

TARGET SCSI / FC
CONTROLLER
HOST SCSI / FC
HBA CONTROLLER
Source: Salomon Smith Barney

Below in Figure 183 are some examples of an HBA and host and target controller
chips from QLogic.
Figure 183. I/O Interfaces Exist in the Form of Chips, Boards, and Cells

QLogic ISP2312 - 2Gb, Dual Channel QLogic FTEC482 - 2Gb Fibre


Fibre Channel Controller Chip Channel HD Controller Chip

QLogic SANblade 2300


2 Gigabit Fibre Channel HBA
Source: QLogic and Salomon Smith Barney

181
The SAN Book III – October 22, 2001

Anatomy of a Fibre Channel HBA


Silicon integration is To understand the different Fibre Channel HBA solutions offered by QLogic,
crucial for long-term Emulex, JNI, and others, we believe a close look at the host bus adapter is
success in this market.
educational. Each of the three leaders develop their own silicon and software, which
are two critical elements of market success. Just as with IDE and SCSI, a portion of
the Fibre Channel interface market will eventually go on the motherboard; therefore,
a silicon integration path is crucial to servicing the lower-end, more price-sensitive
portion of the market.
Figure 184 below details the primary components of an HBA: the Fibre Channel
controller ASIC, memory (usually SRAM), firmware (usually Mask ROM or Flash),
voltage regulator (LDO), serializer/deserializer (SerDes), and the connectors
(Electro-optic Module, standard nine-pin DB for copper, GBIC for fiber optic cable,
or more recently released Small Form factor Plugs).
Figure 184. Dissection of a Fibre Channel Host Bus Adapter

LDO

FLASH

SRAM

OSCILLATOR
QLA2200 ASIC

Note: the SerDes is embedded within the chip

Source: QLogic and Salomon Smith Barney

The HBA Bone Is Connected to the PCI Bone...


The HBA bone is connected to the PCI bone. The PCI bone is connected to the
server backplane bone, which is connected to every other bone. We have illustrated
the elusive PCI (Peripheral Component Interconnect) on the left of Figure 185
below. It is the internal slot in the back of your computer that you plug stuff into,
like modems, extra memory, NIC cards, the connector for your joystick, etc.
Note: The S-Bus slot, depicted on the right below, looks similar to a PCI slot, but is
not (see section below the figure for details).

182
The SAN Book III – October 22, 2001

Figure 185. HBAs Slide into the PCI Slot, Which Is Connected to the Server or PC’s Backplane (Left)

S-Bus Connector
in SPARC System

PCI Connectors
in PC

Source: Salomon Smith Barney

SBus Is the Same, but Not Exactly


SBus was developed by Sun Microsystems in the 1980s to connect graphic cards to
its SPARCstations. It is a 25 MHz, 32- or 64-bit bus using an enclosed pin
connection on the underside of the board. The form factor for the board is smaller
than PCI, about 85mm x 150mm versus 100mm x 180mm (excluding pins, which
are another 10mm at the bottom). Because it only operates at 25 MHz, the
maximum bandwidth is 200 MB (25x64/8) versus 528 MB for PCI.
SBus has recovery capabilities that PCI does not. Therefore, if the card seizes, the
channel can reset itself without taking the system down or “hanging” the system. In
other words, SBus is self healing.
While a PCI HBA stands up on end, an SBus HBA lays flat on top of the
motherboard. In Figure 185 above, if the SBus HBA were plugged in, you would
not be able to see the Super SPARC chip because the HBA would lay (flat) over the
top of it. Below in Figure 186, we have illustrated the physical differences between
PCI and SBus HBAs from JNI.
Currently, Sun (the main supplier of SBus servers) appears to be moving away from
SBus in favor of PCI and PCI-X, but that has not stopped the recent entrants of
QLogic and Emulex into the SBus market. We believe there will be a respectable
legacy SBus market for a while, although at a diminishing rate.
Note: We go into greater detail of PCI and PCI-X in this report’s “Protocols”
section.

183
The SAN Book III – October 22, 2001

Figure 186. JNI PCI and SBus HBAs

SBus Interconnect

PCI Interconnect

Source: JNI

Lean and Mean: QLogic and JNI Similarities


QLogic has the highest Largely driven by exceptional ASIC design teams, QLogic and JNI share design
silicon integration, and philosophies with regard to chip and board designs. Both integrate a RISC
complementary HBA and
chip solutions.
processor, which runs at 33–66 Mhz and is primarily used for command processing,
rather than data processing. Instead, the data path goes from the Fibre Channel
cable, through an integrated FIFO in the ASIC, to the PCI interface. The primary
difference between QLogic and JNI is the integrated SerDes. QLogic’s integration
of the SerDes provides nearly a single chip solution for the HBA. This strategy
allows QLogic to be the only HBA vendor with a complementary chip offering
(which echoes the direction it is moving in the switch market). The company is able
to provide OEMs with either a chip solution for their motherboard, or a board
solution (HBA), both using the same software driver and chipset.
Though QLogic’s Fibre Channel host IC business is small today (about $4–$5
million or 8% of their Fibre Channel revenues), we believe it will provide an
important avenue for growth in the future as OEMs such as Sun Microsystems
choose silicon solutions for lower-end, more price-sensitive products. Already,
other independent makers of HBAs and RAID adapter boards leverage QLogic
silicon, including American Megatrends, ATTO, IBM, and VMIC.
QLogic’s single chip solution positions it as the lowest cost provider, enabling it to
charge the lowest end-user ASP (Average Selling Price) at about $1100 per board
(compared with about $1700 from other vendors) and get the highest gross margins
in the HBA industry (i.e., 64.5% in fiscal 2001 ending March, 2001).
Note: While JNI and QLogic do use GBICs (GigaBit Interface Converters) and
SFPs (Small Form factor Plugs), the vast majority of HBAs are sold with dedicated
optical or copper connections that are hardwired onto the board, thereby, lowering
cost and increasing reliability.

184
The SAN Book III – October 22, 2001

Figure 187. QLogic and JNI Designs Have Similar Data Paths

QLogic Fibre Channel HBA JNI Fibre Channel HBA

GBIC GBIC
OPTIONAL OPTIONAL
TO FIBRE SRAM TO FIBRE SRAM
CHANNEL CHANNEL
CABLE SERDES CABLE
F
F I
I F
F SERDES O
O RISC RISC
PCI DMA

TO PCI BUS TO PCI BUS


Source: Salomon Smith Barney

JNI is shifting from a While JNI has previously based its HBAs on Agilent’s Tachyon ASIC, in order to
licensed ASIC to its gain more control of its product roadmap and lower production costs it is
internally-developed
“Emerald” platform.
transitioning to its own ASIC family called Emerald. Interestingly, the Emerald
technology was originally sold to JNI by Adaptec, which dislodged its Fibre
Channel division in 1998. JNI’s new chip release, Emerald, serves as the basis for
nearly all of its new products, including PCI-X. By owning its own ASIC
architecture, JNI will be better served as the Fibre Channel market evolves to 2Gbps,
dual-channel, and support for Virtual Interface (VI), Internet Protocol (IP), and
iSCSI. In addition, a strong semiconductor design team will be crucial for JNI if it
decides to formulate a chip strategy for future growth.
Emulex Takes the High Road
The Emulex design is not necessarily centered around silicon integration, but around
performance (including reliability and functionality) and software instead. As
shown in Figure 188 below, Emulex uses an Intel StrongARM processor, which runs
up to 200–250 MHz, to manage the data flow. The benefit of Emulex’s design is
that it can better and more quickly leverage industry standard microprocessor
advances. Also, Emulex’s higher chip count circumvents distant limitations and
increases signal integrity (both of which are important in high-end and SAN
environments).
The downside to the chip-intensive solution is that a high part count increases costs
and takes up a large footprint; thereby requiring significant reengineering to migrate
to an embed solution for the motherboard (i.e., a chip solution). The higher part
count also gives Emulex a high-cost structure. As a result, Emulex’s gross margins
trend around 50%, less than QLogic and JNI which have historical gross margins
between 55% and 65%.
Another benefit of Emulex’s design is the software/firmware platform, which as a
result of the standardized hardware design, handles much of the data flow and
control and includes support for Fibre Channel and FICON on every board.
Integrating these functions in software, rather than silicon (as QLogic and JNI have),
provides greater flexibility for Emulex customers. As a result of Emulex’s chip-
intensive solution, it is well respected in the industry as the premium provider.

185
The SAN Book III – October 22, 2001

Figure 188. Emulex LP8000 and Block Diagram Showing Data Path
LDO FLASH
SRAM
Emulex Fibre Channel HBA
StrongARM
LDO

GBIC
TO FIBRE SRAM
CHANNEL
CABLE

DragonFly
SRAM SERDES

StrongARM

SerDes

DragonFly ASIC TO PCI BUS

Source: Salomon Smith Barney

Emulex’s flexible Emulex’s HBA architecture is ideal for high-end OEM customers, who often
architecture is ideal for implement proprietary features in their products. As a result, IBM uses Emulex’s
high-end customers.
products throughout several platforms including Sequent NUMA, RS6000, and
S390. In addition, IBM’s next generation FICON based mainframe system, referred
to as the zSeries (previously code-named Freeway), uses Emulex HBAs in both the
front-end server and back-end Shark storage subsystem. IBM has the only
subsystem that uses an HBA rather than a target chip. This means Emulex gets a
twofer in captive IBM server-to-storage environments by putting an HBA on both
ends of the connection.
In addition, Emulex’s design bug fixes are easily applied in its firmware which is
helpful when problems arise; and bugs are not uncommon. For example, in
September, 2000, Emulex customers discovered a bug in their LP8000 HBA, which
the company fixed and corrected in its software. As a result, shipment of new
product was only put on hold for a week and existing customers downloaded a
simple software update from Emulex’s website. This could have been a much more
costly problem if Emulex used a more integrated design.

High-end...Low-end?
Although we attribute Emulex’s design to the high-end market and QLogic’s to the
volume market, both have been making strides to cover the other’s so-called market.
In the end, it is more about winning customers. The first vendor to win a customer
can establish significant barriers to entry due to server and storage companies’ tight
development and integration to specific HBA designs (be it the high-end or volume
market). Note: It takes a new HBA design many months to get OEM approval, even
from an existing supplier using the same drivers.
Note: While QLogic supplies HBAs into Sun’s high-end products, Emulex supplies
to EMC and IBM’s volume NT and NAS products. Integration by vendor is often
more important than by platform.
The first mover advantage is almost everything in the HBA market. Emulex has
many first-mover design wins, including IBM, EMC, and Compaq. Qlogic has the
first mover advantage at Sun, Dell and NetApp. JNI for EMC (Sbus) and HDS.
Thus far, its has proved to be very difficult (and very unsuccessful) to try unhooking

186
The SAN Book III – October 22, 2001

an incumbent HBA supplier. Clearly this market cannot commoditize until HBAs
from different vendors are seamlessly interchangeable (or “plug and play”) which
we expect to take some time.
The Emerging HBA’int Market: a Single-Chip Solution
QLogic and JNI have the If Fibre Channel follows in the footsteps of the IDE and SCSI interfaces, a new
highest integration levels market segment will emerge: a chip based (silicon) solution for motherboards. We
.
believe this migration is crucial to promote broader Fibre Channel adoptions by
helping it penetrate the lower end of the market that is not as performance sensitive
and is more price sensitive. To date, for the 1 Gbps generation of Fibre Channel
products, QLogic is the supplier with the highest degree of integration with on-board
SerDes and RISC processor in its ASIC; the integration enables QLogic to have the
highest gross margins and the lowest ASPs. This low cost has enabled QLogic to
penetrate the entry-level Windows NT markets.
JNI’s Emerald-IV ASIC also integrates about every part needed to incorporate a
single chip solution, except the SerDes. JNI chose to avoid the complexities of
SerDes integration, recognizing that vendors such as Vitesse sell the 1 Gb part
(which is generally manufactured in GaAs) for only $5.
Trade-offs with an At the end of the day, while single chip solutions might be ideal for some
integrated SerDes. applications, they will not cover the mid- and high-end markets. For example, in our
opinion, signal clarity is highest with an external SerDes and larger SRAM buffers.
Therefore, full-blown SAN applications requiring external cabinet connections often
prefer the cost/performance trade-off of the higher-end HBAs (such as Emulex’s).
However, for smaller SANs and especially NAS applications, which may only
require inter-cabinet storage, we expect a more integrated HBA or chip solution
(such as QLogic’s) to be preferred.

Competitive Landscape
We expect market conditions to become more competitive; providing increased
share to the stronger players (the gorilla game). Already, strong players such as
Emulex, JNI, and QLogic have demonstrated solid growth rates, while smaller
players have underperformed. In addition, traditional SCSI providers, with the
exception of QLogic, have failed to gain traction in the Fibre Channel HBA market.
Below in Figure 189, we plot the various growth trajectories of the three leading
Fibre Channel HBA vendors: Emulex, JNI, and QLogic.

187
The SAN Book III – October 22, 2001

Figure 189. Historical Growth of Fibre Channel HBA Revenues per Quarter
Revenues in millions

$80
$70
$60
$50
$40
$30
$20
$10
$0
Mar-99 Jun-99 Sep-99 Dec-99 Mar-00 Jun-00 Sep-00 Dec-00 Mar-01 Jun-01

Emulex QLogic JNI

Source: Company reports and Salomon Smith Barney

Players’ Positions
The various market positions of each of the Fibre Channel HBA vendors are
revealed by a study of their respective market shares. A recent IDC study of 2000
results (shown in Figure 190 below) proclaimed Emulex the unit and revenue share
leader achieving 32.1% and 34.8% of these respective markets. We attribute
Emulex’s strong showing to its focus on a high-end design solution which spans the
UNIX, NT, and mainframe platforms, translating into a broader footprint and higher
ASPs.
The study also shows that in 2000, QLogic was able to maintain a disproportionate
share of the units shipped when compared to its revenue share. QLogic’s unit and
revenue market shares were 29.1% and 17.0%, respectively. QLogic focuses more
closely on the price/performance market and does not command the same pricing
premium as Emulex. However, it does a similar unit volume and if QLogic’s host
and target chips were included we believe that QLogic would be the unit market
share leader.
JNI has an interesting position in the market: most of its shipments have been for
Solaris systems with about 70% of the company’s revenues from Solaris SBus
adapters. Also, the company’s close relationship with EMC has been lucrative given
the popularity of the Sun server/EMC storage configuration.

188
The SAN Book III – October 22, 2001

Figure 190. Various HBA Companies’ Positions

Unit Share (2000) Revenue Share (2000)


Interphase Other Other
Sun Interphase
3% Sun
3% 2% 3% 2%
6%
Compaq Agilent
7% Emulex 7% Emulex
JNI 32% 35%
12% Compaq
12%

Agilent
12%
Qlogic
Qlogic JNI
17%
29% 18%

Source: IDC (July, 2001)

Agilent’s Tachyon ASIC Agilent was one of the early developers of Fibre Channel technology; however, by
is used in several early our analysis, the company’s success appears somewhat understated in the HBA
generation host
adapters.
market share numbers. The market share data show that Agilent has not had much
success in the HBA market, however, its ASIC (called “Tachyon”) development
efforts landed the company several design wins on several early generation HBAs.
The Tachyon hardware is based on a state machine architecture, which allows
performance of the HBA to scale proportionally with the system CPU.
Below in Figure 191, we illustrate some board and silicon relationships.
Figure 191. Board/Silicon Relationships
ADAPTEC COMPAQ EMULEX INTERPHASE JNI QLOGIC SUN MICRO

SUN
EMULEX
AGILENT AGILENT AGILENT ASIC
DRAGONFLY JNI QLOGIC
TACHYON TACHYON TACHYON or
or CENTAUR EMERALD ISP2x00
ASIC ASIC ASIC QLOGIC
ASIC
ASIC

Source: Salomon Smith Barney

Gaining Share from Captive Suppliers


Both Compaq and Sun Compaq and Sun are the only two systems manufacturers who also produce Fibre
have relationships with Channel HBAs. Even though both started with home-grown solutions in their server
independent HBA and
ASIC vendors.
systems, both now have established relationships with merchant (independent)
suppliers, which we expect will continue to expand in the future. Compaq has
relationships with Emulex for NT and high-end UNIX, JNI for Solaris, and QLogic
for Linux. Meanwhile, Sun has a long-standing relationship with QLogic through its
SCSI platform and has been using QLogic Fibre Channel products (both ASICs and
HBAs) as well. Given the torrid pace of Fibre Channel development, we would
expect Sun and Compaq to migrate toward merchant solutions, rather than continue
to develop their own HBAs internally. By increasingly taking share from vertical

189
The SAN Book III – October 22, 2001

suppliers, independent suppliers such as Emulex, JNI, and QLogic should be able to
easily outpace industry growth rates.
While the independent HBA suppliers (Emulex, JNI and QLogic) significantly
increased their combined market share from 39% in 1998 to 70% in 2000, the
captive suppliers (Compaq and Sun), which declined in market share to 18% in 2000
from 53% in 1998, have clearly begun to pull out.
Figure 192. Market Share of FC Host Bust Adapter Suppliers

1998 2000
Other Interphase
Interphase Agilent Sun Other
0% 3%
8% 0% 6% 2%
Qlogic
Agilent
JNI 9%
7% Emulex
9% Sun 35%
Compaq
43%
12%

Compaq
10%
Qlogic
Emulex JNI
17%
21% 18%

Source: IDC (July, 2001)

Name Association
Because the Fibre Channel HBA resides within a server, and ultimately connects to
a storage subsystem, it is a critical determinant of the overall system performance.
Therefore, end users generally select the HBA that has been approved by either the
server vendor, such as Compaq and Sun, or, the storage vendor, such as EMC,
Compaq StorageWorks, or Hitachi Data — or both! To make matters more
complex, the HBA can also be procured either from the server vendor, storage
vendor, or through indirect distribution channels. The following figure illustrates
some supplier/customer relationships for Emulex, QLogic, and JNI.

190
The SAN Book III – October 22, 2001

Figure 193. HBA Supplier/Customer Relationships

IBM (all), EMC (all &


NAS), Compaq (UNIX

Emulex
OEM 85% & NT) , HDS (all),
Fujitsu (Solaris),
McData (all)
Bell, Info-X, Tech Data,
Channel 15% AVNET, Tidalwire, ACAL

QLogic Sun (Solaris), Dell


(NT), NetApp (NAS),
OEM 90% IBM (NT), SGI, EMC
(NT), Compaq (Linux)

Channel 10% Bell, Info-X, Tech Data, Consan

EMC (Solaris), HDS (UNIX


OEM 76% & NT), Compaq (Solaris),
IBM (UNIX & NT),
JNI

StorageTek (Solaris & NT)


Bell Micro, GE Access,
Channel 24% ACAL, Info-X, Tidalwire,
NetMarks
Source: Salomon Smith Barney

Which relationships are most important? In our view, clearly, the largest server and
storage OEMs are most influential. One popular combination has been to combine
Sun Solaris servers, NT servers from Compaq or Dell, and EMC storage units, all
connected using a Brocade switch. Resellers would likely recommend a JNI HBA
be used for the Sun server, since EMC has tested and qualified JNI’s SBus HBA.
Sun and other vendors may offer an HBA, but unless it has been certified by EMC
also, the use of it may void the EMC warranties. Needless to say, Compaq, IBM,
Dell, MTI, and other major OEMs have similar programs.

Sun’s Approach Is a Bit Tricky


When it’s Sun server-to-Sun storage, Sun sells a QLogic HBA. When it’s a Sun
server -to-someone else’s storage, a JNI HBA is used and can be bought from the
“other” storage vendor or from the channel (distributors).

Platform Focus Determines Product Strategy


UNIX (excluding Solaris)
Emulex’s flexible Emulex has captured most of the Fibre Channel HBA market for high-end UNIX
platform is ideal for high- platforms through its close relationship with IBM, EMC and, recently, Compaq.
end UNIX server OEMs.
Emulex drivers support several flavors of UNIX including AIX, VMS, OS/390
UNIX, Sequent’s Dynix, HP-UX, and DG-UX. Furthermore, as IBM mainframes
adopt Fibre Channel through its FICON (FIbre channel CONnect) protocol, Emulex
will enter the mainframe market on both the host side as well as in IBM’s FICON
Shark storage subsystem (a twofer). Linux drivers are available from all three major
HBA vendors including Emulex, QLogic, and JNI. Note: with IBM’s AIX and NT

191
The SAN Book III – October 22, 2001

storage sales, Emulex often gets a twofer, since it also embeds an HBA in IBM’s
Shark in those environments.
Below we have drawn a rough sketch of Emulex’s platform penetrations based on
our own in-depth industry research. Note: A good portion of Emulex’s revenues
come from the storage side within Shark which we have assigned to the AIX and NT
platforms. In the future, FICON should emerge as a noticeable contributor for
Emulex. We have based our estimates on HBA unit volumes.
Figure 194. Emulex’s Platform Profile

NAS
Tru64
1%
4%

AIX
31% NT
46%

Solaris
18%

Source: Salomon Smith Barney

Solaris
JNI has captured most of Sun Micro’s version of UNIX, called Solaris, is used mainly on Sun and Fujitsu
the HBA market for (which licenses a version of Solaris from Sun) servers. Emulex, JNI, QLogic, and
Solaris SBus, which is
commonly used in Sun’s
Sun each have Solaris drivers available for HBAs built for the PCI bus, which is
high-end servers. used in entry-level and workgroup servers from Sun.
Emulex, JNI, Sun and (recently) QLogic each have Solaris drivers available for
SBus HBAs, though JNI is recognized as the leading provider of SBus HBAs for
Solaris. QLogic has historically had the closest relationship with Sun including
some embedded chip design wins. We expect QLogic to continue to further
penetrate Sun.
Below in Figures 195 and 196, we have drawn a rough sketch of QLogic and JNI’s
respective platform penetrations based on our own industry research. We have
based our estimates on HBA unit volumes.

192
The SAN Book III – October 22, 2001

Figure 195. QLogic’s Platform Profile

NAS
Solaris 9%
12%

NT
79%

Source: .Salomon Smith Barney

Figure 196. JNI’s Platform Profile

AIX & HP-UX


5%
NT
5%

Solaris
90%

Source: Salomon Smith Barney

Windows NT/2000
QLogic supplies major QLogic is the predominant supplier of Fibre Channel HBAs for the high-volume
NT server vendors such Windows NT server market. QLogic’s mini-port driver, which is commonly used
as Dell and IBM.
for NT systems, is highly regarded in the industry. As such, QLogic has captured
much market share in this segment and has relationships with OEMs such as Dell
and IBM’s Netfinity group. Adaptec provides HBAs to Hewlett-Packard for its
NTserver platforms, while Emulex is a supplier to Compaq for its Proliant NT
servers.

193
The SAN Book III – October 22, 2001

Storage Networking Equipment


➤ We view storage networking equipment as the highest octane portion
of the storage industry. We estimate that SANs are only 5%–10%
penetrated, on their way to about 90%. This should enable many
players, including subsystem and software vendors, to grab their fair
share of the market’s growth.

➤ We believe the storage networking market presents an awesome


growth opportunity and still has its best days in front of it.

The Storage Networking Equipment Market


We expect SAN adoption We define the storage networking equipment market (which we refer to simply as
to lead the storage “storage networking”) to include: bridges, routers, hubs, switches, and Directors (not
industry into a new
phase of evolution.
including HBAs). We will mostly focus on switches and Directors.
Regardless of IT demand in 2001, storage networking companies experienced
phenomenal growth in 2000 — 347% for Brocade, 178.5% for Inrange, and 161.1%
for McData (according to the companies’ filings). Dataquest expects the storage
networking market to grow to $12,630 million in 2005 from $760 million in 2000,
representing a 75.4% CAGR.
Figure 197. Storage Networking Growth (2000–05)
Revenues in millions

$14,000
$12,000 75.4% CAGR
$10,000
$8,000
$6,000
$4,000
$2,000
$0
2000 2001 2002 2003 2004 2005

Source: Dataquest and Salomon Smith Barney

What Is a SAN Switch?


To really take full To really take full advantage of the speed and scalability of Fibre Channel, a fabric
advantage of the speed switch must be deployed. A switch provides any-to-any (point-to-point)
and scalability of Fibre
Channel, a fabric switch
connectivity, thereby enabling each port to communicate at full capacity (currently
must be deployed. 1 Gbps or 2 Gbps).

194
The SAN Book III – October 22, 2001

Figure 198. SAN Switch Diagram

Fibre Fibre
Server Channel Channel

Switch
Server
Source: Salomon Smith Barney

FC-AL Versus Fabric


A switch can be FC-AL or fabric. There are many differences between FC-AL
switches and fabric switches. A FC-AL switch can be thought of as an upgraded
hub. In other words, it acts like a hub from an addressing standpoint, but moves data
like a switch at full bandwidth per port (review hubs below). The difference in
addressing reduces a FC-AL switch’s available connectivity. Typical FC-AL
switches can connect up to 32,000 nodes, while fabric switches can connect up to 16
million nodes (theoretically). The increased connectivity capabilities of a fabric
switch are why it is referred to as the foundation of today’s SAN.
One of the advantages of One of the advantages of FC-AL is that some legacy host bus adapters (HBAs) use
FC-AL is that some the FC-AL protocol, not the fabric protocol. This means that a software translation
legacy host bus adapters
(HBAs) use the FC-AL
has to be performed in a fabric switch to connect to legacy FC-AL HBAs. This is
protocol. embedded (for a price) into a switch called an FL-Port, to emulate the FC-AL
protocol and enable fabric-to-FC-AL connectivity. (HBAs are storage network
interfacing cards that plug into servers to enable SCSI or Fibre Channel
connectivity, much the same way a Network Interface Card [NIC] does in the LAN.)
Another way to interconnect fabric switches to FC-AL HBAs is through embedded
software. In this architecture, software is added to a native fabric switch (which
would fit into the device administration segment of storage software) to, in effect,
dumb it down (remember, a FC-AL device is less complex than a fabric device).
The additional software makes the fabric switch think and act as if it were a FC-AL
switch in order to interconnect it with legacy FC-AL HBAs. Since this is an
embedded software-driven solution, upgrading back to full-fabric simply requires
paying a software licensing fee and flipping a software switch. This is a relatively
seamless process that does not cause any network interruption. (This is the
architecture of Brocade’s and QLogic’s FC-AL switches.)
Alternatively rather than dumbing down the entire switch, Brocade and other
companies also have the ability to do reverse translation which allows legacy FC-AL
servers to be translated onto the fabric and utilize fabric storage or tape devices. The

195
The SAN Book III – October 22, 2001

fabric communications are translated back to the legacy server for a seamless two-
way communication. Since only the setup is done in software (actually firmware),
once loaded, the translation occurs in the ASIC hardware and performance can be
sustained at full Fibre Channel speeds.
Gadzoox has approached this in a unique way by developing its Capellix chassis-
based switch. The chassis part means that it is modular in order to plug either a FC-
AL switching blade or fabric switching blade into a slot, thus making the switch
perform with the desired protocol. This solution requires that the system be taken
down (including unscrewing the lid of the switch) in order to upgrade. In our
opinion, this process is just about the last thing an IT manager wants to do.
Switch Addressing — Public Versus Private Loops
Hubs and FC-AL Hubs and FC-AL switches use simple local addressing schemes, much like an office
switches use simple phone system. To dial a person across the floor, you key in four digits. You don’t
local addressing
schemes. Fabric
have to dial the whole number because you’re in a local network. These devices are
switches and Directors referred to as “private loops,” because they can only see and communicate with
use more advanced other devices within their private loop.
(public) addressing
technology. Fabric switches use a more robust addressing scheme. These devices can be thought
of as your telephone at home. To transmit your voice, you dial the three-digit area
code, the three-digit zone, and then the four-digit address. This tells the phone
switch what area to go to, what zone to go to, and then what exact address to go to
within that zone. This is the same way a SAN fabric switch operates. Fabric
switches are referred to as “public” because they can see and communicate with
most every device connected to the network.

Switching Architectures
We believe that software Different vendors have architected their switching products with varying
remains one of the most components, cost structures, reliability, scalability, flexibility, and sophistication.
important features when
assessing the
While switching hardware architectures are wildly different (and we’ll go through
competitiveness of the range), we believe that software remains one of the most important features
assorted products. when assessing the competitiveness of assorted products. Software leverages
hardware to provide customers with solutions. We believe McData’s recent
acquisition of SAN Navigator (a storage software company focused on network
management) is an outcrop of the demand for software functionality. It has been our
observation that the best solution, not the best hardware, grabs the most market
share.
Switching Interconnects
Similar to our discussion of storage subsystem backplanes and PCI bus versus
Infiband, a switched interconnect can significantly increase performance. Using a
crossbar switch architecture, however, demands more logic, which adds complexity,
increases costs and potentially delays a product’s time to market.
Below in Figure 199 we illustrate a few different switch (or Director) architectures,
including Bus (Brocade and QLogic), switched (McData), multi-stage (Inrange), and
passive (Brocade’s Silkworm 12000).

196
The SAN Book III – October 22, 2001

Figure 199. Switch Interconnects

4 port ASIC 4 port ASIC


Bus Interconnect
Cross Bar Switch

ASIC ASIC

ASIC ASIC

4 port ASIC 4 port ASIC


Multi-stage Interconnect Passive Interconnect

ASIC ASIC

ASIC

ASIC ASIC

Source: Salomon Smith Barney

Brocade
Brocade uses a bus interconnect and leads the market in product maturity and ASIC
functionality. Brocade’s developments — trunking, time stamping, frame filtering
and traffic monitoring — are among the industry’s leading functions. Brocade’s
focus on real, end-user needs and a “total solution” design has enabled it to maintain
its popularity with end users.

Brocade Design Snapshot:


➤ bus interconnect
➤ eight-port ASIC design (2 Gbps)
➤ distributed memory
➤ distributed controller
Figure 200 below illustrates Brocade’s 2 Gbps 16-port switch architecture.
Note: The shaded region represents the ASIC’s footprint.

197
The SAN Book III – October 22, 2001

Figure 200. Brocade’s 2Gbps 16-port Switch Architecture

RS-232
10/100 Ethernet

P ow
er
SFP
From Fibre
Channel
Buffer
Cable SerDes

(send)

Pentium
Class CPU

High Speed Proprietary Bus Interconnect


SerDes

Controller DRAM

Flash

SerDes

SFP

To Fibre
Channel
Cable
(receive) SerDes
Memory
Buffer

Source: Salomon Smith Barney

McData
McData takes the highroad with a serial crossbar interconnect. The serial crossbar
enables any-to-any port connectivity without performance degradation. Since
McData uses a serial crossbar interconnect it must traverse two SerDes (or the same
one twice) per every send or receive transmission (or four per every send and receive
transmission). The extra SerDes workload is brought on due to the “serial”
technology within the crossbar which requires a serial transmission versus the ASIC
which requires a parallel transmission. For this reason, McData uses two SerDes per
ASIC (each SerDes used by McData has four quad channels). The SerDes (which
stands for Serializer/Deserializer) transforms the signal into a serial transmission
(Serializer) and/or into a parallel transmission (Deserializer).
McData uses a serial crossbar interconnect in order to increase performance and
utilize off-the-shelf technology (in this case, from Vitesse).
McData has developed a four-port ASIC design and has the highest component
count in the industry, which means it also has the highest cost structure. The benefit
is that McData also has the most reliable design. If one of its ASICs fails, it only

198
The SAN Book III – October 22, 2001

takes down four ports, versus higher ASIC densities, which can take down as many
as 16 ports upon failure.
Below in Figure 201, we have laid out a 16-port McData switch (ES-3016) and
mapped a data send and receive transmission in order to make sense of this
gibberish.
Figure 201. McData’s 2 Gbps 16-port Switch Architecture

RS-232
10/100 Ethernet
Buffer
Pow
SFP er
From Fibre
Channel Flash
SerDes
Cable SerDes
(send)
DRAM

Controller Intel I960


CPU
Serial
SerDes
SerDes Crossbar
Chip

SFP
To Fibre
Channel Serial
SerDes
Cable SerDes Crossbar
(receive) Controller

SerDes
SerDes

McData’s 4-
port Viper
ASIC

Source: Salomon Smith Barney

McData Design Snapshot:


➤ serial crossbar interconnect
➤ four-port ASIC (1 and 2 Gbps) with serial crossbar interconnect
➤ memory is distributed and integrated
➤ distributed and integrated controller
QLogic
QLogic’s exceptional ASIC design team has focussed on ASIC integration (i.e., chip
design), resulting in an ASIC with the greatest integrated functionality and
connectivity in the industry. QLogic’s chip integrates so many components that its
16-port switches (called SANbox) require only a single chip. QLogic has even

199
The SAN Book III – October 22, 2001

embedded crossbar technology inside its new Itasca Fibre Channel chip. As with its
HBAs, this provides QLogic with the most compact and lowest cost design in the
industry.

QLogic Design Snapshot:


➤ single-chip solution
➤ 16 ports per ASIC (2 Gbps)
➤ integrates the controller, memory, buffer, and crossbar inside the ASIC
Figure 202. QLogic’s 2Gbps 16-port Switch Architecture

Pow
er RS-232
10/100 Ethernet

SFP
From Fibre SerDes
Channel
Cable
(send)
SerDes
GL_Port 0 Pentium
Class CPU

SerDes

Controller
DRAM
SerDes

Flash

GL_Port 11

Interconnect
Embedded
SerDes QLogic’s
16-Port
Itasca
SerDes
ASIC
SFP
To Fibre
Channel
Cable SerDes
(receive)

SerDes

Source: Salomon Smith Barney

The Core of a Switch/Director Is its ASIC


Although switch Although switch architectures can impact performance and connectivity, ASIC
architectures can impact densities are also important to track. McData uses a four-port ASIC architecture,
performance and
connectivity, ASIC
Brocade uses 8 ports (four ports in its 1 Gbps ASIC), and QLogic/Inrange use 16
densities are also ports (8 ports in the 1Gbps ASIC). Increasing the number of ports per ASIC can
important to track. significantly reduce costs and increase performance; however, it also inherits greater

200
The SAN Book III – October 22, 2001

port loss upon failures. While this might not be an issue for lower- and mid-range
environments, higher-end applications might prefer lower port densities per ASIC
(not to be confused with ports per switch or Director, they are very different
concepts that do not run parallel). Lower ASIC port densities are often viewed as
more reliable because if an ASIC does fail, it takes down a smaller fraction of the
overall switch or Director (i.e., if an ASIC fails, only four ports fail rather than eight
or 16). That said, ASIC failures are very unusual and redundancies can be built in
for greater reliability.
Other important aspects Other important aspects of ASICs are their functionality and software integration.
of ASICs are their While a customer might not pay as much attention to port count per ASIC, they
functionality and
software integration.
surely recognize the delivery of functionality. Functionality and software
integration are key elements to the successful delivery of a solution (i.e., what’s best
for the customer). For example, the ability of Brocade’s ASIC to autosense FC-AL
and fabric environments has enabled it to grab the lion’s share of the storage
networking market early on, as customers migrate from FC-AL to fabric. The
inability of other vendors to effectively mix these environments could have stalled
these vendors’ attempts to penetrate the market.
Other valuable, upcoming ASIC features include trunking, time stamping, frame
filtering, and traffic monitoring.

Trunking
Trunking, combined with load balancing, is one of the more anticipated inter-switch
functions. Trunking aggregates bandwidth to create a single, larger pipe that all data
can traverse. Load balancing equalizes the distribution of data across multiple
channels. Combining the two features results in better performance and less
congestion.
Below in Figure 203, we illustrate the way Trunking works when combined with
load balancing.

201
The SAN Book III – October 22, 2001

Figure 203. Trunking Functionality

Optimal bandwidth utilization using load balancing


on up to four 2G links

Without Trunking
2G 1G
2G
1.5G Congestion 1.5G
.5G .5G
1G 1G
2G 1G

2G With Trunking 2G
1.5G 1.5G
.5G .5G
1G 1G
8Gbps ISL Trunk:
2G All streams share bandwidth
2G
ASIC preserves
in-order frame delivery
Source: Brocade

Frame Filtering
Frame filtering is another ASIC capability that would integrate with available
software and enable features that would allow for successful system deployment.
With frame filtering, many features can be developed to improve management and
monitoring capabilities, and security in the SAN. Features such as LUN zoning,
detailed performance analysis, and multi-protocol zoning could improve the SAN
deployment and overall management, as well as ensure future integration with other
protocols.

Switch Market and Competitive Landscape


Switch revenues are expected to grow to $5,970 million in 2005 from $470 million
in 2000, representing a 66.3% CAGR, according to Dataquest.

202
The SAN Book III – October 22, 2001

Figure 204. Switch Market Growth


Revenues in millions ($)

$6,000

$5,000 Switches 66.3% CAGR

$4,000

$3,000

$2,000

$1,000

$0
2000 2001 2002 2003 2004 2005

Source: Dataquest

The Biggest Little Kid on the Block


The switch market not The switch market not only has the largest share within storage networking by
only has the largest revenue, it is also the largest by ports shipped. This latter point is of particular note
share within storage
networking by revenue, it
given that Directors had much higher port densities in 2000 (eight and 16 ports for
is also the largest by switches compared to 32 and 64 ports for Directors). Detailed analysis, therefore,
ports shipped. points to many more “switch” installations than the port shipment differential
depicts. Note: Although the hub port market share might appear impressive, its
actually down approximately 12% from 1999, while switch port shipments were up
approximately 317.3% from 1999, according to IDC.
Figure 205. Switch Storage Networking Market Share

2000 Market Share (Revenue) 2000 Market Share (Ports)

Hubs Directors
Directors 7% 10% Hubs
27%
31%

Switches Switches
66% 59%

Source: Dataquest (left) & IDC (right)

Players
Brocade is the clear Brocade is the clear leader in the Fibre Channel switch segment, garnering an 86%
leader in the Fibre market share in 2000, according to IDC. Brocade has been able to maintain its
Channel switch segment.
switch market share at around 90% since the industry’s inception. We expect
McData, a long-time Director-only supplier, to play a greater role in the switch

203
The SAN Book III – October 22, 2001

market since the release of its 16- and 32-port switches, which began shipping at the
end of 2000, and were adopted by OEMs early in 2001. QLogic should also play a
bigger role in the future with its Sun relationship, early delivery of 2 Gbps and
integrated chip design, which should prove to be very leverageable in the embedded
switch market. Note: QLogic is the only vendor in both the Fibre Channel switch
and HBA markets.
Figure 206. Switch Market Share (Revenues) 2000

Vixel Gadzoox
QLogic 2%
3%
9%

Brocade
86%

Source: IDC (July, 2001)

Brocade Remains the Leader


From a revenue point of From a revenue point of view, it cannot be disputed that Brocade continues to be the
view, it cannot be leader of the storage networking market. As illustrated below in Figure 207, all
disputed that Brocade is
the leader of the storage
companies have chalked up some impressive growth. The Figure charts Brocade’s,
networking market. Inrange’s, QLogic’s and McData’s switch and Director revenue growth (per
company filings and our own research).
Figure 207. Switch and Director Revenue Growth — Brocade, McData, Inrange, and QLogic
Revenues in millions ($)

$180
$160
$140
$120
$100
$80
$60
$40
$20
$0
Mar-99 Jun-99 Sep-99 Dec-99 Mar-00 Jun-00 Sep-00 Dec-00 Mar-01 Jun-01

Brocade McData Inrange Qlogic

Source: Company reports and Salomon Smith Barney

204
The SAN Book III – October 22, 2001

Below in Figure 208 we highlight Brocade’s Silkworm product portfolio.


Figure 208. Brocade’s Silkworm Product Family

Silkworm 12000
Silkworm 6400
(64-port core switch)
(Pre-configured
Silkworm 2400 64-port switch)
(8-port switch)

Silkworm 2800
(16-port switch)

Source: Brocade and Salomon Smith Barney

Below in Figure 209, we also illustrate QLogic’s switch portfolio, including its
Director, which it OEMs from Inrange. QLogic has been successfully growing the
switch business it bought with its acquisition of Ancor in August, 2000. In
QLogic’s June 2001 quarter its switch business grew, to $14.4 million, or about 10%
sequentially, up from the $7–$9 million level when QLogic acquired it.

205
The SAN Book III – October 22, 2001

Figure 209. QLogic’s Switch and Director Product Family

Source: QLogic and Salomon Smith Barney

What Is a Director?
Think of Directors as Defining a Director is like defining the perfect cheeseburger: It depends on who you
high-end, fully redundant ask. By definition, Directors are high-end storage switches.
switches.
We have found that the most common criteria for Directors include:
➤ high port density
➤ scalability
➤ fully redundant, fabric, non-blocking architecture that is hot swapable
It is important to keep in mind that there are two very different types of storage
Directors: ESCON/FICON and Fibre Channel.
ESCON and FICON Directors (mainframe)
ESCON (mainframe) ESCON (mainframe) Directors have been in use for almost a decade. ESCON
Directors have been in Directors provide mainframe connectivity; in other words, in order to connect
use for almost a decade.
multiple storage subsystems to a mainframe one must use an ESCON Director.
Mainframes communicate through ESCON (Enterprise System CONnection), much
the same way that open system servers and storage subsystems communicate using
SCSI.

206
The SAN Book III – October 22, 2001

FICON (S/390 FIbre Channel CONnectivity) is to mainframe environments what


Fibre Channel is to open systems (UNIX and NT): It’s a more advanced form of
connectivity. We expect Inrange and McData to be significant players in the FICON
Director market.
Figure 210. ESCON and FICON Directors

Transfer Rates Max Distance Throughput


ESCON 17MBps 3 kilometers Half Duplex
FICON 100MBps 20 kilometers Full Duplex
Source: Salomon Smith Barney

ESCON/FICON Directors have been built for zero downtime (or five nines of
reliability, meaning 99.999% uptime). In order to accomplish this goal, Directors
have been designed with multiple hardware redundancies, including power supplies,
fans, backplanes, processors, and memory, as well as the software functionality (i.e.,
logic) to utilize its fully redundant architecture in case of component failures.
Industry sources indicate that McData had an approximate 85% share of the ESCON
market in 1999 by supplying IBM’s 9032 Director that scales to 248 ports per unit.
Inrange captured about 15% of the market with its CD9000 that scales to 256 ports
per unit.
Note: IBM’s FICON products launched in the fall 2001.
Fibre Channel Directors (Open Systems: UNIX and NT)
Inrange and McData Since Inrange and McData (a former subsidiary of EMC) already had
entered the Fibre ESCON/FICON expertise, when they entered the open systems storage market, both
Channel market with
Director designs.
companies approached it from their Director heritage; thereby creating a new
category of Fibre Channel switches called Directors. (Directors are typically about
twice as expensive as Fibre Channel switches on a per-port basis; switch ASPs are
about $1300 per port; Director ASPs are about $2500–$3000 per port). McData’s
first generation 32-port Fibre Channel Directors (the ED-5000) began shipping in
quantity in October, 1998, and Inrange introduced its 64-port Director (the FC/9000)
in April, 2000.
Director...Switch...So what the Heck Is the Difference?
Thus far, the market has essentially defined Directors as high port density switches.
McData’s Directors had 32 ports; Inrange’s 64 ports. Directors, however, are much
more than simply high port densities: They are fully hardware redundant Fibre
Channel fabric switching devices that can maintain full non-blocking bandwidth in
the event of multiple hardware failures. In contrast, switches would have to
substantially over-allocate ports in order to ensure a non-blocking, fully available
and reliable architecture. But, the resulting switch mesh would result in a higher
cost and a more complex, harder to manage environment.
Baby You Can Drive My Car
We liken the technology to a car with four tires going down the highway at 60 mph.
Then one tire blows out (i.e., hardware failure). The Director drops down a
replacement tire and keeps on trucking at 60 mph uninterrupted (like a James Bond

207
The SAN Book III – October 22, 2001

car). The software-redundant switch, on the other hand, redistributes (balances) the
weight of the car such that it can run on the remaining three tires until the fourth tire
can be replaced. This type of switching architecture might suffer some performance
degradation and have to reduce its speed in order to maintain total throughput.
Now imagine that the car has 64 or 128 tires. One tire blowing out is a lot less
significant. In other words, the larger the fabric, the less important one port (tire)
becomes. However, while we believe a software redundant “core switch” might be
able to service a good portion of the market, in our opinion, many IT managers will
simply demand hardware redundancies in order to be conservative. That said, we
believe that switches (including 16-port switches) will continue to dominate the
market, from the point of view of total ports shipped, for at least another couple of
years. As Brocade often points out and we agree, the issue will continue to be about
the size of your overall fabric and how its managed.

Director Market and Competitive Landscape


Directors are projected Director revenues are expected to grow to $6,240 million in 2005 from $190 million
to be the highest growth in 2000, representing a 101.0% CAGR, according to Dataquest. Directors are
area in storage
networking on a
projected to be the highest growth area in storage networking on a percentage basis,
percentage basis. as illustrated in Figure 211 below.
Figure 211. Directors Are Expected to Achieve the Highest Growth in Storage Networking
Revenues in millions ($)

$7,000 Router 48.4% CAGR


$6,000 Hub 3.7% CAGR
$5,000
Switch 66.3% CAGR
$4,000
Director 101.0% CAGR
$3,000
$2,000
$1,000
$0
2000 2001 2002 2003 2004 2005

Routers Hubs Switches Directors

Source: Dataquest

Players
In 2000, McData was the clear Director leader with an 85.4% market share.
However, Inrange definitely deserves the “up-and-comer” award for establishing a
14.6% share after only having product available in the second half of the year.

208
The SAN Book III – October 22, 2001

Figure 212. Director Market Share (Revenues) 2000

Inrange
15%

McData
85%

Source: IDC (July, 2001)

Inrange
Inrange entered the Fibre Channel Director market in April, 2000, with its 64-port
FC/9000 (the largest density at the time) and has since scaled its Director up to 128
ports (the largest density in the market). Inrange uses QLogic’s (formerly Ancor’s)
Fibre Channel ASIC technology in its Director and this has created a tight
partnership between the two companies. Inrange OEMs QLogic’s eight- and 16-port
switches and QLogic OEMs Inrange’s FC/9000 Director.
Inrange has also delivered its own ESCON and FICON Directors, which it sells
through its direct sales force, resellers, and OEMs. The company has also stated its
intention to migrate to a 256-port Director by 2002 (current 64- and 128-port
FC/9000s are architected to be fully upgradable to 256 full bandwidth, non-blocking
ports). Inrange can offer Fibre Channel (fabric and FC-AL, both public and private)
and FICON connectivity within a single box. The FC/9000 also supports IP-over-
Fibre Channel.

McData
McData entered the Fibre Channel market in the fourth quarter of 1998 with its 32-
port ED-5000 Director, which used third-party ASIC technology. In November,
2000, McData delivered its own ASIC technology and introduced its 16-port ES-
3016 switch, its 32-port ES-3032 (March, 2001) and its 64-port ED-6064 (April,
2001). McData also manufactures and designs its own line of ESCON Directors,
ESCON-to-FICON bridge cards (which fit into its ESCON Director), and FICON
Directors.
McData’s new generation 16-, 32- and 64-port products all support Fibre Channel
fabric and FICON in the same enclosure and provide the industry’s most densely
packaged solution (i.e., offering the most ports per square inch). Note: McData does
not support FC-AL in its switches or Directors, except in its ES-1000, which works
in FC-AL mode but not in fabric mode.

Brocade
Brocade is planning to enter the Director market with its Silkworm 12000 late this
year. The Silkworm 12000 is a 64- and 128-port “core switch” that provides much

209
The SAN Book III – October 22, 2001

of the same reliability as a Director though it does not, however, support FICON
(i.e., the mainframe). We expect the Silkworm 12000 to drive significant growth for
Brocade in 2002 and beyond as it opens up the Director market and enables the
company to expand its footprint.
The figure below gives a rough timeline of the release of many Fibre Channel
switching technologies.
Figure 213. Fibre Channel Product Introductions

Brocade &
2Gbps
QLogic Inrange McData

Brocade to
McData 16- introduce 64-port
Brocade 8/16-port port ES-3016 Silkworm 12000
Silkworm
2400/2800
Brocade 64-port
Silkworm 6400

1999 2000 2001 2002

Inrange 64-port Inrange 128-port


FC/9000 FC/9000
QLogic 8/16-port Inrange to introduce
SANbox 256-port FC/9000
McData 32-port
McData 32-port ES-3032 & 64-
ED-5000 port ED-6064

Source: Salomon Smith Barney

Figure 214 presents Inrange’s complete product portfolio.

210
The SAN Book III – October 22, 2001

Figure 214. Inrange’s Product Portfolio


Channel Extender
DWDM FC over WAN
9801SNS
FC/9000
VL/9000
(128 ports)

FC/9000
(64 ports)

FC/9000
(16 ports)

Source: Inrange

We Believe High Port Density Is The Sweet Spot


We see higher port We believe that higher port density switches and Directors are one of the next big
density switches and waves of demand currently unfolding in storage networking. In fact, we believe 128
Directors as one of the
next big waves of
ports (perhaps even up to 256 ports in a single box) is going to be the ultimate sweet
demand. spot. Note: Legacy mainframe ESCON Directors have about 256 ports.

Size Does Matter! Why Bigger Is Better


We believe the industry is moving to higher port density products as fast as they are
delivered. Why? Firstly, higher port densities offer scalability. For example, if an
IT manager only needs 24 ports today, they can purchase a FC/9000 populated with
only 24 ports. The important aspect of the FC/9000 is that it can scale to 128 ports
as the customer’s demands increase. We equate this to the way IT departments use
storage subsystems. With storage subsystems, if an IT department needs half of a
terabyte of storage, they usually buy a storage subsystem that scales to 3 TBs or
more and only populate enough disk drives to satisfy current demand. This provides
seamless scalability in order to easily satisfy future demand without forklifting in,
installing, and configuring a new subsystem every time additional capacity is
needed. With the FC/9000, capacity can easily be added by snapping in additional
ports (up to 128 ports and 256 ports in the future).
Higher port count switches also enable more efficient scalability. As multiple
switches are interconnected to create a fabric or network, ports must be sacrificed as
e-ports to provide inter-switching links. This, in turn, creates a loss of ports as
fabrics are scaled. By utilizing higher port count products, fewer ports are lost and
the overall network remains less complex, due to the interconnectivity of fewer
switches/Directors.
Below in Figure 215 we illustrate how using 16-port switches to build a fully
redundant non-blocking 32-node fabric requires six switches (96 ports). Then, in
Figure 216 we illustrate a more common architecture which scales to reach 30

211
The SAN Book III – October 22, 2001

nodes, but only uses three 16-port switches (48 ports). However, the latter example
is not a fully redundant, non-blocking architecture, meaning that performance can
degrade as port failures occur.
Note: The diagrams below are also illustrations of a multistage or meshed
architecture.
Figure 215. Building a 32-port High Availability Fabric Using 16-port Switches

16-port Switches
8 nodes

8 nodes
8 nodes
8 nodes

*Note: there are less expensive/complex ways to build a 32-port fabric. However, the less
expensive solution will not be as robust. We architected the above design to compare apples to
apples on the high-end without sacrificing performance and availability.

Source: Salomon Smith Barney

Figure 216. Building a Simple 30 node Fabric Using 16-port Switches

16-port Switches
10 nodes

10 nodes
10 nodes

*Note: we believe this is the most common multi-switch fabric


architecture, but it is not a fully redundant, non-blocking fabric.

Source: Salomon Smith Barney

212
The SAN Book III – October 22, 2001

16-port Switches Will Likely Stay In Demand for Years


We believe that low port We believe that low port density switches (eight, 16 and 32 ports) will continue to
density switches will be in high demand for many more years: Not everyone wants, or can afford, to build
continue to be in high
demand for many more
a large-scale storage network. We also expect to see low port density switches
years. continue to thrive on the edge of the network and in department locations for many
more years. We believe that SANs are still in the early stages of deployment, which
we feel will translate into demand for lower-scale, lower-cost solutions that enable
phase one deployments. However, longer term, we believe the low port density
market will stagnate and, eventually, contract.
One architecture many industry experts expect to continue gaining momentum is
core-to-edge architecture. In this architecture, a core switch or Director would sit in
the middle of a storage fabric and smaller, less expensive switches would line the
periphery (as illustrated in Figure 217 below).
Figure 217. Core-to-Edge Storage Networking Architecture

Storage
Servers

Switches Switches

Directors or
Core Switches

Core

Edge Edge

Source: Salomon Smith Barney

Thus far, McData has been the leader in the core-to-edge market due to its time to
market delivery of a full solution spanning Directors and switches. However, we
expect Brocade and Inrange to establish solid footholds in this segment.
Figure 218 shows a more likely core-to-edge architecture.

213
The SAN Book III – October 22, 2001

Figure 218. More Likely Core-to-Edge Storage Networking Architecture

Edge
Core

Source: Salomon Smith Barney

Below in Figure 219 we illustrate McData’s core-to-edge product portfolio.


Figure 219. McData’s Core-to-Edge Product Portfolio

ES-1000 ES-3016

ES-3032
ED-6064
Source: McData

Compatibility Matters!
Interoperability is a key Although many products seem robust, when you get into actual computing
component to promoting environments involving multiple vendors and platforms, they fall a bit short due to
broad SAN adoption.
incompatibility. In other words, what good is a switch that cannot connect to

214
The SAN Book III – October 22, 2001

Microsoft’s Windows2000? Or to Sun’s Solaris UNIX platform? Or what good is a


switch that cannot connect to EMC’s Symmetrix storage subsystem? Or to
Compaq’s StorageWorks? Or to an HBA? We believe this is an important issue to
overcome in order to promote broader SAN adoption.
Perhaps one of the more pressing issues is inter-switch vendor connectivity.

Brocade Is The Standard


In June, 2000, Brocade announced that its Fabric Shortest Path First (FSPF) routing
protocol was accepted by the Technical Committee T11 of the National Committee
for Information Technology Standards (NCITS), for approval as an industry
standard by the American National Standards Institute (ANSI), one of the primary
Fibre Channel standards bodies.
Brocade submitted the standards proposal in March, 2000, seeking to facilitate
baseline interoperability among Fibre Channel switches from multiple vendors. The
proposal, FSPF, specified a common method for routing and moving data through
various Fibre Channel networking vendors’ switches. We view this as having been
a powerful move for Brocade and a necessity for the industry.
Brocade has also gotten its e-port technology accepted as the inter-switching link
standard. While Brocade’s FSPF is the routing protocol, Brocade’s e-port specifies
the hardware connection.

Interconnecting Vendors’ Switches Is Important


One of the largest One of the largest promises of Storage Area Networks (SANs) has been any-to-any
promises of Storage connectivity. The gating factor prohibiting any-to-any connectivity in Fibre Channel
Area Networks (SANs)
has been any-to-any
networks has been incompatibility between various networking vendors’
connectivity. technologies. By using Brocade’s new inter-operative standards, any-to-any
connectivity can become a reality which, we believe, will serve to increase adoption
rates. But, vendors still have to agree to work together.
Inter-operability Promotes the Industry
We believe Fibre Channel’s recent evolution toward standardization will prove to be
a great ally for storage networking companies by increasing the rate of adoption. It
has been our observation that end users do not like to be locked into proprietary
solutions and that they are more willing to embrace new technologies when
standards exist.
Note: Cisco has also singed on to use Brocade’s e-port technology to interconnect
its Ethernet switches to Brocade’s Fibre Channel fabrics.
We believe 2001
company-specific What Is a SAN Bridge?
momentum will continue. The first obstacle of any new technology is backward compatibility. No IT manager
is going to implement a SAN, or any other technology, if it calls for the disposal of
previous investments. To enable older legacy SCSI systems to inter-operate with
Fibre Channel, a “bridge” must be built (SCSI devices cannot be directly connected
to Fibre Channel devices). Thus, bridges — one of the first SAN networking
markets to develop in early 1999 (the other was hubs, which we will discuss later)

215
The SAN Book III – October 22, 2001

— are an integral part of bringing the concept of Fibre Channel SANs to reality
today.
Bridges can be in stand-alone boxes or integrated within storage subsystems (and in
which case are called blades). Figure 220 shows a SAN bridge.
Figure 220. SAN Bridge Diagram

Fibre Channel SCSI


Disk
Array SAN Bridges
enable SCSI to
Fibre Channel
connectivity.

Bridge
Source: Salomon Smith Barney

In the future, we expect to also see Fibre Channel-to-Ethernet, Fibre Channel-to-


Infiniband, and Ethernet-to-Infiniband bridges in order to migrate today’s
technologies into tomorrow’s environment.

What Is a Traditional SAN Router?


To take a bridge one step To take a bridge one step further, routers enable several SCSI-to-Fibre Channel
further, routers enable connections. Each router has a varying number of SCSI and Fibre Channel ports,
several SCSI-to-Fibre
Channel connections.
depending on the vendor and product. One of the benefits of routers is that they
look at the data being sent through them. Where hubs and switches focus on moving
data frames quickly and efficiently, a router stops the data frame (for a microsecond)
to determine if it’s a read, write, or copy command and routes it accordingly.
(Stopping data in this instance should not be confused with a disruptive event; this is
intentional and does not interrupt data transfers). The router determines what needs
to be done with the data frame by reading a leader on each frame of data.
As a result of being able to read a frame of information without ruining a backup,
routers (or the router function) have the potential to play an increasing and
influential role in future SANs in order to enable, say, serverless backup.
Much like bridges, routers can be in stand-alone boxes or integrated (as blades)
within storage subsystems. We project the blade market, which has just begun to
emerge, to offer decent growth for the next couple of years. It provides storage
vendors, which do not have Fibre Channel connectivity integrated into their
subsystem, the ability to offer a plug-and-play, Fibre Channel–ready subsystem
without substantially changing their subsystem’s architecture. (That is, it makes a
legacy SCSI tape library appear to have native Fibre Channel connectivity. In other
words, you can plug a Fibre Channel cable directly into a SCSI tape library via an
embedded routing blade.)
Furthermore, as Fibre Channel migrates to incorporate LAN (Ethernet) and WAN
(ATM) interconnectivity, we believe the router’s place in SAN networks will be

216
The SAN Book III – October 22, 2001

cemented. Note: In Dataquest’s analysis and projections, bridges and routers are
placed in the same category.
Figure 221. SAN Router Diagram

Fibre Channel SCSI


Disk
Array SAN Routers
enable multiple
SCSI to Fibre
Channel
connections.
Router
Tape
Library
Source: Salomon Smith Barney

Players
Crossroads has been the Crossroads has been the clear leader in the SAN router segment, although there are
clear leader in the SAN other emerging niche companies entering the market, such as Chaparral, ATTO and
router segment.
Pathlight (recently bought by ADIC).
Figure 222. Crossroads’ Product Portfolio

Crossroads 8000
Front View

Rear View

Source: Crossroads

Upcoming SAN Routing


We believe storage As SANs begin to expand and require inter-SAN connectivity, we believe current
routing could become a networking technologies will be leveraged. In other words, SANs will be
very large portion of
overall storage
interconnected over Ethernet and ATM. We also believe Dense Wave Division
networking revenues. Multiplexing (DWDM) will play a big role for storage vendors (which is why

217
The SAN Book III – October 22, 2001

companies such as Brocade, Inrange, and McData have partnered up with leading
DWDM companies OpticalNetworks and ADVA, respectively).
Figure 223. Routers Could Enable Multiple Protocol Connectivity

Fibre
Ethernet Channel
Disk
Array SAN Routers can
enable the
interconnectivity
of multiple
protocols.
Router
Tape
Library
Source: Salomon Smith Barney

We believe storage routing could become a very large portion of overall storage
networking revenues (although it may become tough to distinguish this segment as
storage per se). We believe this market will gain traction in 2002 to provide SAN to
LAN, WAN, and MAN connectivity.

218
The SAN Book III – October 22, 2001

Figure 224. SAN-to-SAN Interconnectivity via a LAN

FC-GE
Router
Server 1 Disk 1

SAN
Disk 2
Server 2

Tape 1
LAN Server 3

Disk 3

Server 4 Disk 4

Disk 5
Server 5

SAN
Disk 6

Server 6

NAS Disk 7
Server 1

NAS
FC-GE
Server 2 Disk 8
Router

Source: Salomon Smith Barney

Players
Cisco entered into a technology license agreement with Brocade in June, 2000.
Under the terms of the agreement Cisco’s Catalyst 6000 product family of
multilayer switching technology (Ethernet) would be combined with Brocade’s
Silkworm switch technology (Fibre Channel). The interconnection would occur
through the integration of a Fibre Channel–based interface, designed by Brocade and
built by Cisco, located within a Cisco switch, providing switch-to-switch
connectivity. The combined solution is based on encapsulating Fibre Channel’s
protocol, FCP, over TCP stacked with IP through Ethernet to achieve non-blocking
OC48 performance (about 248 MBps). Cisco plans to continue increasing the speed
of transmission with future generations of the interface.

219
The SAN Book III – October 22, 2001

In our opinion, other leading companies in the SAN router segment include ADVA
(through its purchase of SAN Ltd.), Crossroads, Computer Network Technologies,
Nishan, and SAN Valley.
Figure 225. Nishan’s FC-to-IP Router

Nishan IPS 3000

Source: Nishan

Figure 226. SAN Valley’s FC-to-IP Router

SAN Valley Systems SL1000

Source: SAN Valley

What Is a SAN (Entry Level) Hub?


One promise of a SAN is any-to-any connectivity. This can be achieved with a
“hub.” Hubs were one of the first SAN networking market segments to generate
revenues. A hub connects several storage subsystems and allows them to
communicate with one another. This would be like a walkie-talkie: Anyone can
speak to anyone else. The drawback of hubs is that data must share bandwidth; in
other words, only one person can speak at a time.
Hubs are constructed using a shared bandwidth architecture. A hub can transmit
data at 100 megabytes per second (MBps), which is also the speed of Fibre Channel,
but it must allocate that bandwidth such that the sum of all ports capacity usage
equals 100 MBps. In other words, only two ports in a hub can be connected to each
other and be transmitting data at one time; therefore, various ports must take turns.
This inefficient use of bandwidth has led to the migration to more sophisticated
devices: switches.

220
The SAN Book III – October 22, 2001

Figure 227. SAN Hub Diagram

Fibre Fibre
Server Channel Channel

Hub
Server
Source: Salomon Smith Barney

What Is a SAN Managed Hub?


Managed hubs not only connect various storage subsystems, they also manage them.
One of the taboos in computer storage is data interruption. Data interruption can
cause a backup to fail. Failed backups lead to lost data. A managed hub can
overcome some of the disruptive events that occur in an entry hub (also called a non-
managed hub or dumb hub), by controlling port traffic.
Disruptive events result in delays in data transmission. These delays can result from
a new storage device connecting to the hub for the first time or from the failure of an
existing connection. In both cases, entry hubs inadvertently delay data transmission
while they reconfigure their environment. A few milliseconds of disruption in a data
transmission can result in a failed backup, even if the backup is eight hours into the
process (in which case the backup might have to start over).
Another benefit of managed hubs is that they can act as an initiator. For example,
who tells storage subsystems when it’s time to back up? Or what to back up? A
managed hub can be the initiator of a data transfer. Therefore, managed hubs can be
a real enabler for serverless and LANless backup.
Players
We continue to believe the hub market, be it entry level or managed, is losing its
luster as storage networks demand more bandwidth and greater connectivity. In
fact, many of the “hub” vendors, such as Vixel and Gadzoox, have been trying to
migrate up to the switching segment. However, both companies have had difficulty
managing through the transition.

221
The SAN Book III – October 22, 2001

Figure 228. Gadzoox’s Product Portfolio

Gibraltar GS FC_AL hub

Capellix 3000 Modular SAN switch

Capellix 2000 SAN switch

Slingshot 4218 2Gb Open Fabric switch

Slingshot 4210 2Gb Open Fabric switch


Source: Gadzoox

Hubs, We Don’t Need No Stinking Hubs


In our view, the hub market is all but gone. Although we believe hubs will continue
to play a small role in the low-end and along the outer periphery of fabric SANs, in
our view, it is a commodity market and has very little (if any) potential for future
growth.

Hub and Switch Side-By-Side Comparison


Hubs can only transmit data at 100 MBps for the entire device, compared with
switches that can transfer data at 100 MBps per port (note: as Fibre Channel
transmission speeds increase, these capacities will also increase). For example, a
four-port hub can transfer 100 MBps (200 MBps if you take into account bi-
directional data transmission), while a four-port switch can transfer 200 MBps (400
MBps if you take into account bi-directional data transmission). Note: each link
takes up two ports.
The reason for the difference in data transfer rates can be found in the architectures.
In Figures 229 and 230 we illustrate hub and switch architectures.

222
The SAN Book III – October 22, 2001

Figure 229. Four-port Hub and Switch Architectures

Hub Switch

Source: Salomon Smith Barney

Figure 230. Six-port Hub and Switch Architectures (16-port Switches Are More Common)

Hub Switch

Source: Salomon Smith Barney

Note: The two previous figures were drawn for simplicity; in reality, any port in a
switch can connect to any other port. In our illustrations, switch ports only connect
across.
Notice that all the hub ports are connected to each other (called daisy chaining). In
order for one port to connect to another port in a hub, it must pass through all other
ports along the way. In a switch, various paths are available to simultaneously
connect multiple ports without interfering with one another creating point-to-point
connectivity.

223
The SAN Book III – October 22, 2001

Ever Hear of a Looplet?


Switches can also use “looplets” to expand connectivity. Looplets act in much the
same way as an entry hub and can connect a maximum of 126 nodes. A switch with
multiple looplets is illustrated in Figure 231 below.
Figure 231. Fibre Channel Arbitrated Loop (FC-AL) Diagram

Looplet 1

Switch
Looplet 2

Hub or
Switch

Source: Salomon Smith Barney

What’s the Difference?


Notice that Looplet 1 simply connects devices to one another to form a loop (called
daisy chaining), while Looplet 2 uses a hub or switch to make the connections. The
advantage to using a hub or switch, rather than daisy chaining, is it makes wiring the
loop less cumbersome, and if a device on the loop fails, the loop can still function.
Daisy chaining devices act much like Christmas tree lights; when one bulb goes out,
they all go out.
However, we project hubs have already become commodities, while switches have,
and will continue to demand, premium prices as a result of their more complex and
leveragable technology. For this reason, hub manufacturers have been attempting to
increase their value by increasing the functionality of their product set (i.e., they are
moving to switches).

224
The SAN Book III – October 22, 2001

Protocols

225
The SAN Book III – October 22, 2001

z
Protocols: The Language of SANs
➤ While Fibre Channel continues to proliferate the industry in the form
of networked and non-networked storage, IP storage is getting closer
to becoming a reality. Several IP based products have been
announced recently, and the IETF is getting closer to selecting a final
standard. We expect IP storage to effectively penetrate the market
over the next three years, but instead of replacing Fibre Channel, we
believe it will complement it on the low end and at the periphery in
the near term.

➤ We also anticipate InfiniBand showing up over the next few years as


the next evolutionary development to PCI. InfiniBand delivers a
switch-based I/O technology which could, eventually, be leveraged
into the storage network fabric, potentially competing against both
Fibre Channel and IP for the holy grail of storage networking.

Multiple Technologies Are Peering into Each Other’s


Markets for Future Growth
We maintain our belief Of course every company wants to take its technology as far as it can go; however,
that Fibre Channel will in the past, various networking technologies have been locked into their respective
continue to be the only
robust storage
markets due to technology barriers: IP for LANs, ATM for WANs, and Fibre
networking technology Channel for SANs. Figure 232 below illustrates how the various networking
for another two to three technologies are beginning to bump heads: Everyone wants to expand into each
years.
other’s markets.
Note: We have drawn the overlaps where we believe they exist today. We expect
these overlaps to expand over time and a LAN/SAN overlap to emerge.
Figure 232. Various Networking Areas

WAN

LAN
SAN

Source: Salomon Smith Barney

226
The SAN Book III – October 22, 2001

On top of the seeming On top of the seeming sibling rivalries, newer protocols continue to show up.
sibling rivalries, newer InfiniBand appears one of the most promising. Over the next three years we are
protocols continue to
show up.
looking forward to watching InfiniBand’s evolution as a PCI bus replacement (i.e.,
an alternative I/O) and as a potential storage networking contender. However, the
most proliferated and aggressive technology has been IP (i.e., Ethernet; IP stands for
Internet Protocol), which has already crept into the WAN space by providing voice
over IP. IP is now positioning itself, through the recent introduction of iSCSI and
other iterations, to also enter storage networking.
Before diving into a protocol discussion, let’s first appreciate why so many different
vendors, technologies and investors are so intensely focused on this debate: Storage
networking is going to be a HUGE market and everyone wants a piece of it.
In this section we explore the various technologies and their respective potentials in
the storage market. In the process we will look to provide detailed explanations of
each technology and how we expect them to play out.
➤ Basics of I/O Interfaces – starts out in the weeds of system and connectivity
performance (I/O stands for Input/Output).
Various protocols can be used to network storage. We discuss 1) Fibre Channel, 2)
Ethernet, and 3) IP storage:
➤ Fibre Channel: Networking Storage — outlines the importance of Fibre
Channel and its advantages in storage networking.
➤ Fibre Channel versus Ethernet — Compares Fibre Channel and Ethernet and
tracks the histories of both.
➤ IP Storage — Comparisons of iSCSI, iFCP and FCIP, three IP-based protocols
gaining traction for storage networking with backing from influential companies
such as Cisco and IBM.
➤ InfiniBand — a new architecture that multiplies the maximum bandwidth of
interconnectivity in servers.

The Reality of a Homogenous Network


We believe one of the We believe one of the most misunderstood assumptions is that end users want, or
most misunderstood need, one ubiquitous networking technology to be the silver bullet that solves all
assumptions is that end
users want, or need, one
problems for all computing needs so that one homogenous networking ring can be
ubiquitous networking installed to handle all forms of network traffic equally as well.
technology.
If that were true, why do we have many different flavors of UNIX, or use
mainframes, UNIX, and NT together in the same data center (such as Citigroup’s)?
Why are there three different types of disk drives (ATA, SCSI, and Fibre Channel)?
Why are there bicycles, mopeds, and motorcycles? Why trucks and cars? Why cats
and dogs? Why boys and girls? Okay, we are getting a bit carried away, but you get
the point.
The reality is that we want choices and certain technologies solve certain needs
better than other technologies, but that is not going to stop people from trying to
deliver the perfect ubiquitous network technology.

227
The SAN Book III – October 22, 2001

When?
In the end, we believe it We have heard many people implying that this concept of “one network” is just
will take at least two or around the corner and beginning to emerge today. While we agree it is being
three years before
Ethernet is ready to go
architected today, we believe the availability of a technology that can meet all types
head-to-head with Fibre of networking demand is in the distant future (more than five or ten years away).
Channel. Thereafter, we
believe the two It has been our experience that technology shifts of this magnitude take more time
technologies will — not less — than originally anticipated. Right now, people are talking about
overlap. Gigabit Ethernet moving to 10 GbE and 40 GbE; we note that Ethernet is currently
ramping transmission speeds of 1 Gbps. At the same time, it must be noted that
Fibre Channel is not standing still. 2 Gbps Fibre Channel has already hit General
Availability (GA) and the 10 Gbps Fibre Channel spec is already in the works.
In Figure 233 below we have depicted our expectations of the rollout of various
storage protocols.
Figure 233. Past and Potential Future Storage Networking Technology Availability
Adoption

el
nn

e
ag
ha

nd
tor
C

iba
S
re

fin
IP
Fib

In
1998 1999 2000 2001 2002 2003 2004 2005 2006

Source: Salomon Smith Barney (August, 2000)

How?
The how is a bit tougher to explain. In the following sections we will explain the
difference between various networking protocols and what we believe will unfold in
the next three years. While a longer-term vision is exciting to talk about, we believe
it is too futuristic and difficult to predict and, therefore, should not consume
investors’ diligence today. Besides, the technologies will probably change at least
15 times in the next couple of years alone.
Learning from History
Who would have guessed that Cisco’s technology — a company that very few
people had heard of in the late eighties and early nineties — could have usurped Big
Blue’s efforts to network the Internet with its technology? The point is, nobody
really knows what technologies and companies will survive in the distant future (we
believe a more appropriate window to focus on is two or three years out).

228
The SAN Book III – October 22, 2001

Who Wins?
We certainly see many new technological advancements developing from all sides
We believe when the of the debate. For example, we believe Cisco’s torrid pace of acquisitions, including
dust settles in ten years, NuSpeed, is a clear indication that it recognizes the constantly changing network
we will have networking topology. We also believe almost every one of the Fibre Channel companies has
companies with Ethernet
roots and ones with
morphed quite a bit since inception. For example, Ancor, which was focused on
Fibre Channel roots. In Fibre Channel LANs in the early nineties, moved into Fibre Channel storage
our view, the key for networking, was bought buy QLogic, and is now developing iSCSI and InfiniBand.
investors is not to pick a Another good example is Inrange which has morphed from a telecom and datacom
technology, rather pick
quality companies which
networking company with some mainframe products to be a leader in the high-end
have displayed the Fibre Channel market. Inrange has long-haul IP and ATM connectivity and is
ability and willingness to looking to incorporate InfiniBand and iSCSI in future architectures.
adapt.
We believe the ultimate winners will be companies from both Ethernet and Fibre
Channel networking ancestries that have displayed 1) technology leadership, 2)
visionary management leadership, 3) the ability to react quickly to changing
technologies and changing customer demands, and 4) a willingness to evolve their
core competencies to suit customer needs two to three years out by developing fresh
technologies in new directions today.
Our Thoughts for the Future
➤ We believe the emerging “Data-centric” architecture (architecture, not network)
will incorporate multiple networking platforms, including Fibre Channel,
Ethernet, ATM, DWDM, and InfiniBand.
➤ Traditional networking companies will both compete and partner with Fibre
Channel networking companies well into the future.
➤ Ethernet will develop into a more robust technology over the next three years,
and only then, will it be able to begin competing head-to-head with Fibre
Channel to network storage.
➤ 10 Gbps Ethernet (10 GbE) will tip the scale toward using Ethernet, perhaps in
the form of iSCSI, as the preferred solution in the low end of the market and on
the periphery of storage networks.
➤ By the time Ethernet has developed the same level of functionality as Fibre
Channel, it will no longer be TCP/IP Ethernet. We believe it will require a
forklift upgrade (i.e., the installation of an entirely new IP network) with
hardware that can communicate with the new protocol, which we expect to be
different from today’s Ethernet hardware. We believe that iSCSI has taken the
lead as the most viable IP storage networking protocol, but expect that others
will continue to evolve as challengers.
➤ Fibre Channel will continue to be the unchallenged, premier storage networking
solution for enterprise class, production data for at least another two or three
years.
➤ Fibre Channel will continue to gain momentum through 2004, no matter how
many resources are stacked against it. Note: It takes a long time between a

229
The SAN Book III – October 22, 2001

technology’s inception and when it’s ready for prime time. It took Fibre
Channel, for example, more than five years to get ready for prime time.
➤ Storage Networking companies should continue to recognize record growth and
storage networking will likely expand into a HUGE market opportunity.
➤ InfiniBand should replace the PCI bus over the next two to three years and only
then will it begin to show up as a potential alternative to network storage.
➤ The Virtual Interface (VI) protocol will play an increasing role as systems get
more advanced and require faster I/O.
➤ Storage Networking companies have already begun and will continue to
incorporate Ethernet, ATM, InfiniBand, and any other networking technology
which presents itself as a viable storage networking alternative for the future.

230
The SAN Book III – October 22, 2001

The Basics of I/O Interfaces


➤ The change from a SCSI interface as a server-to-storage interconnect
to Fibre Channel has made large-scale SANs possible. Knowing the
basics of I/O interfacing can help investors evaluate new
connectivity technologies such as IP storage and InfiniBand.

Connectivity Speeds SAN Adoption


Connectivity from server to storage has been an area where some of the greatest
strides have been made and continue to be made. The change from a Small
Computer System Interface (SCSI) to a Fibre Channel interface has made large-scale
SANs possible.
In this section, we explain the basics of Input/Output (I/O) systems, both past and
present, to gain a better understanding of how the future of interconnectivity
standards might emerge and how this might affect storage networking.

What Is I/O?
Input/Output (I/O) I/O stands for Input/Output and describes the process and components involved in
describes the process sending and receiving data between computers and peripherals. For example,
and components
involved in sending and
computers can use the SCSI interface for its I/O with disk drives. SCSI is a protocol
receiving data between describing one way of sending and receiving data. Alternatively, computers can also
computers and use Fibre Channel or IDE.
peripherals.
Each protocol has its advantages and disadvantages. Hence, different ones are used
for different purposes. For example, while SCSI and Fibre Channel are used for
communication with disk drives, Universal Serial Bus (USB) protocol is often used
with communication for keyboards.

Critical for System Performance


High-performance computer systems need to be built with high-performance parts.
Low-end “free PCs” and high-end servers are both dependent on four critical sub-
systems: 1) microprocessors, 2) memory, 3) networks, and 4) system I/O — all of
which need to be performing in balance. In other words, the system is only as fast as
its slowest part. The weakest link is often referred to as the “bottleneck”.
The often forgotten link Over time, Intel has successfully conditioned us to view the microprocessor as a
in the chain is the I/O critical determinant of a system’s performance. Also, with the controversy over
infrastructure.
DDR, RDRAM and PC-133 DRAM technologies, the importance of the memory
structure became apparent. A third important determinant of performance was
identified by Sun Microsystems who said “the network is the computer.” As such,
we are now shifting from Ethernet to Gigabit Ethernet within the LAN connection.
The fourth, and less publicized link in the chain is the system’s I/O. Broadly
speaking, this is how a computer transmits data onto and off of the motherboard,
whether it be to a disk drive, tape drive, scanner, printer, etc. This is the place where
the cable meets the computer. Since boosting the I/O transmit rate can enhance
system performance, I/O technologies such as IDE, SCSI (pronounced skuzzy), and

231
The SAN Book III – October 22, 2001

Fibre Channel have each evolved through several generations of performance,


reliability, and functionality.

Cirque de Standards
Because of Intel’s significant presence in microprocessors, chipsets, and
motherboards, it has emerged as a pseudo-governor of computer architecture,
endorsing specific memory or graphics standards to ensure the PC “evolves” in an
orderly manner. To a degree, I/O standards have also been influenced by Intel. Its
endorsement of USB has shifted the low-performance peripherals market away from
serial and parallel ports. High-performance I/O, however, has been out of the
company’s jurisdiction, since Intel only determines specifications up to the PCI bus,
the portal through which many peripherals are attached. As to what devices are
attached to the PCI bus, it is the IT professional or systems engineer that determines
whether the I/O protocol will be IDE, SCSI, or Fibre Channel.

IDE: Absorbed into Chips


In the early 1980s, the Integrated Drive Electronics (IDE) interface, formally known
as Advanced Technology Attachment (ATA), was developed by Western Digital and
Compaq, and installed as the primary link between a computer’s peripheral bus and
a disk drive. Over the years, as bus bandwidths increased, the IDE standard also
evolved into its current form called UltraATA 100, which runs at 100 MBps; the
next generation will be Serial ATA expect to eventually run at 600 MBps. As IDE
became pervasive and Intel’s chipsets began to absorb the IDE interface, the market
for discrete host and peripheral IDE products declined rapidly. Today, almost every
host chipset and every target low-end disk drive ASIC include variants of IDE. The
current generation of IDE, shipping in volume today is UltraATA 100; newer
chipsets, such as Intel’s 845, 850, and 860 have the interface embedded.
IDE will likely remain a The IDE interface will likely remain the primary desktop/workstation interconnect
desktop PC standard. standard because it doesn’t scale as well as SCSI or Fibre Channel technologies.
Only two devices can be connected to a single IDE channel versus 15 for SCSI. A
common desktop configuration has one disk drive and a CD-ROM. If more
peripherals are needed (such as a second disk drive), another interface channel
would be required. Performance workstations and servers often have two to five
peripherals in each system, and require a greater degree of connectivity.

SCSI: the Incumbent Interconnect for Servers


Similar to IDE, the SCSI standard has served primarily as an interface for storage
devices for over a decade. In addition to greater connectivity, SCSI has a
performance advantage over IDE systems. IDE’s newest generation, UltraATA 100,
runs at 100 MBps, while SCSI’s latest generation, Ultra320, runs at 320 MBps.
Finally, the disk drive vendors, in an effort to segment markets, have typically
produced faster, higher-density disk drives with a SCSI interface. SCSI is used
primarily in mid- and high-end computing applications.
Versus IDE, SCSI Because SCSI is not often integrated within a chipset, there is usually additional cost
solutions carry a cost associated with SCSI technology over IDE. A SCSI chip may be a $30–$40 cost
premium.
added on a motherboard, while a SCSI host bus adapter (HBA) may cost $100–

232
The SAN Book III – October 22, 2001

$300. In addition, disk drive manufacturers also charge a premium for SCSI drives.
To justify the cost premium, SCSI vendors have touted both speed and functionality.
The latest generation, Ultra320, has not only the highest data transfer rate of any
SCSI generation, but also new fault tolerance features such as Cyclic Redundancy
Checking (CRC) and Domain Validation. Both improve the integrity and reliability
of the data transfers. With Ultra320, SCSI has managed to keep pace with the
performance of current high-end computer systems (see Figure 234).
Figure 234. Evolution of SCSI

640

320 Future SCSI


Generations
160 Ultra160 SCSI

80 Ultra2 SCSI
MB/sec

LVD
40 Ultra Wide SCSI

20 Ultra SCSI

10 Fast SCSI

SCSI
5

1984 ‘86 ‘88 ‘90 ‘92 ‘94 ‘96 ‘98 2000 2001

Source: Adaptec and Salomon Smith Barney

A Two-front War Limits SCSI’s Growth


UltraATA threatens SCSI Only a few years ago, SCSI was pervasive in high-performance servers and
on low-end and workstations, despite the price premium ($100–$300 for SCSI HBA plus
performance systems.
incremental cost of a SCSI disk drive). Often, the fastest spinning, high density disk
drives were only available with the SCSI interface. In comparison, low-end and
mainstream desktops used IDE drives, with slower spindle rates. Today, SCSI’s
value proposition is not quite as compelling since drive manufacturers have made
IDE drives available with spindle rates up to 7,200 rpm and densities of 40 GB per
platter, nearly as powerful as SCSI and Fibre Channel drives. In addition, on August
29, 2001, a consortium consisting of Seagate, APT, Dell, IBM, Intel, and Maxtor
announced the release of Serial ATA which will eventually transfer data at 600
Mbps. These products are expected to be available in the first half of 2002 with an
initial release of 150 Mbps bandwidth.
Fibre Channel threatens At the other end of the spectrum, in mainstream and high-end servers, SCSI has
SCSI in high-end historically been preferred. High-end systems typically require an array of drives
applications.
(three to nine drives) and longer cable lengths for external storage connections;
SCSI’s ability to connect up to 16 devices per channel at distances up to 12 meters
addressed both of these issues. But recently, the Fibre Channel interface, which can
scale up to 126 devices per channel at distances up to ten kilometers, has garnered
significant interest for high-end systems. Additionally, the Fibre Channel interface
is a cornerstone technology in emerging SAN architectures, thus giving any Fibre
Channel installments even more flexibility and headroom for growth.
Don’t pronounce SCSI With both low-end (IDE) and high-end (Fibre Channel) threats, some industry
dead just yet. analysts have all but declared SCSI a dead technology. But in our opinion, SCSI

233
The SAN Book III – October 22, 2001

remains a reliable and widely available interconnect technology, which has evolved
both in speed and connectivity over several years. Even though Fibre Channel is
gaining traction in the high-end applications, and IDE in low-end, the transition
won’t happen overnight. SCSI’s long history is difficult to overlook due to the fact
that engineers often go with what they have experience with.
For example, EMC has stayed with SCSI drives in its Symmetrix subsystem.
Adaptec, QLogic, and LSI Logic provide SCSI chips and HBAs.
Figure 235. Data Transfer Rates of Various Interface Standards

MBps

350
300
250
200
150
100
50
0 SI

SI
0

0
94

1G

2G
0
1.

2.

10

SC

SC
13
SB

SB

el

el
A

60

20
nn

nn
T
U

aA

a1

a3
ha

ha
eC

eC
ltr

ltr
ltr

U
U

br

br
Fi

Fi

Source: Salomon Smith Barney

Fibre Channel Components Hit the Disk Drive


Disk drive suppliers and component manufacturers have realized the benefits that
Fibre Channel components offer. Today, Fibre Channel technology is not only used
for connectivity between servers and storage, it is also becoming a preferred
component technology. The benefit of Fibre Channel components are 1) their ability
to provide thinner internal connectivity within systems, thereby enabling smaller
systems, 2) they produce less heat, and 3) they enable faster transmissions. For
example, Network Appliance was able to triple its Filers’ capacities to 1.4 terabytes,
simply by using Fibre Channel drives instead of SCSI drives. Subsystems vendors
such as Hitachi Data Systems also realized significant improvements in performance
when they switched from SCSI to Fibre Channel drives.

Alternative Interconnect Technologies Show Promise


USB is becoming a Although IDE, SCSI, and Fibre Channel will likely remain the three primary I/O
ubiquitous low-end interface technologies for storage connections, other technologies have evolved for
solution, 1394 finds
niche in digital video.
other types of devices. For example, the Apple-backed IEEE-1394 (a.k.a. Firewire),
which was once heralded as a competing storage interface, has found acceptance in
digital video applications. Its success is a result of support from both Apple as well
as Sony, which ships a branded version of 1394 called “iLink.” In addition, USB

234
The SAN Book III – October 22, 2001

1.0 has already become the next generation standard for low bandwidth I/O devices
(i.e., keyboards, mice). Already incorporated in current Intel chipsets, USB 1.0 has
lower data transfer rates than either IDE or SCSI and runs at 1.5 MBps. USB 2.0
will have significantly higher performance, up to 60 MBps. While USB 2.0 may
encroach on low-end storage solutions, such as desktop removable media (Iomega
Zip drives), it will not have the bandwidth to compete with the mainstream desktop
I/O standards.

InfiniBand
While this would be a great place to move to a discussion on InfiniBand, we feel it is
more appropriate (chronologically) to spend some time on Fibre Channel next, then
IP storage, and then InfiniBand.

235
The SAN Book III – October 22, 2001

Fibre Channel: Networking Storage


➤ Using Fibre Channel instead of SCSI has evolved storage into a
networked architecture.

➤ The growth and benefits of networking storage are undeniable and


we expect it to present a HUGE growth opportunity as far as the eye
can see!

“Heeeere I Come to Save the Daaaaay!” — Fibre Channel


The existing storage connectivity standard, SCSI, has begun to fall short of the
rapidly advancing system I/O demands. Due to the inherent limitations of the SCSI
standard, a new storage connectivity standard, Fibre Channel, has emerged to take
its place. This new standard has gained strong momentum in the past two years as
the heir apparent to SCSI and the enabler of storage networking (i.e. SANs);
thereby bringing the additional benefits of networking to the storage world.
In this section, we will explain how SCSI evolved to Fibre Channel and why.

Think of It as Super SCSI


Fibre Channel is based Fibre Channel is based on the core SCSI protocol — think of it as souped-up SCSI.
on the core SCSI Fibre Channel improved upon SCSI by morphing it from a parallel to a serial
protocol — think of it as
souped-up SCSI.
interface, allowing Fibre Channel to overcome SCSI’s distance limitations, among
other things.
SCSI is a method for computers to communicate with disk drives. It uses a parallel
interface and its own SCSI protocol to send and receive data. A parallel interface
sends data eight bits concurrently down eight different wires (although the wires are
placed side-by-side to form a single cable). Next to each of those wires is another
wire that describes what type of data is being sent (in Figure 236 below, we label
this type of information “Control”). Note: In reality, the number of wires varies
depending on the version of SCSI being used, there are generally less control than
data lines, and after each control bit several data bits are sent within a frame.
Figure 236. Parallel Versus Serial Interface

SCSI’s Parallel Interface

Data 1
Control 1
Data 2 Multiple
Control 2 Wires
Data 3
Control 3

Fibre Channel’s Serial Interface


One
Data 3 Control 3 Data 2 Control 2 Data 1 Control 1
wire

Source: Salomon Smith Barney

236
The SAN Book III – October 22, 2001

As one can imagine, it is difficult having 16 different pieces of data arrive at the
exact same time. SCSI has distance limitations: The longer the data transfer, the
higher the probability it will arrive at different times. Additionally, having many
electrical wires tightly wrapped together creates electromagnetic interference, which
can degrade the integrity of the data. Fibre Channel overcomes this distance
limitation by using a serial, instead of a parallel, interface. As a result, information
(still using the SCSI protocol) can be sent one bit at a time so the bits of data would
not have a problem in arriving in the exact same order in which they were sent,
regardless of distance.
Additionally, the architects of Fibre Channel added new features into the protocol
which extended its ability to be networked.

Fibre Channel Has Several Additional Advantages to


SCSI
In addition to Few will dispute that Fibre Channel is a compelling technology, even though we
overcoming distance have only begun to see its potential. In addition to overcoming distance limitations,
limitations, Fibre
Channel improves upon
Fibre Channel improves upon many other SCSI shortcomings such as speed, greater
many other SCSI connectivity, centralized management, low overhead, increased reliability, and
shortcomings. improved distance and cabling.
In Figure 237 and the section following we compare SCSI and Fibre Channel.
Figure 237. SCSI and Fibre Channel Comparisons

Maximum Data Maximum Number of


Transfer Rate Maximum Cabling Host Device
(MBps) Length (meters) Connections
Wide Ultra SCSI 40 1.5 16
Wide Ultra2 SCSI 80 12 16
Ultra 160 SCSI 160 12 16
Ultra 320 SCSI 320 12 16
1 Gbps Fibre Channel 100 10,000 126 (loop)
16 mil (switched)
2 Gbps Fibre Channel 200 10,000 126 (loop)
16 mil (switched)
Source: Adaptec and Salomon Smith Barney

➤ High Bandwidth (Speed). The first generation of Fibre Channel debuted with
data transfer rates of 1 Gbps (or 1 Gbit/sec or 100 MB/sec or 100 MBps; it’s all
basically the same thing), faster than any other interface technology at the time.
Currently Fibre Channel’s second generation is being deployed and reaches
twice the data transfer rate, or 2 Gbps (or 200 MBps). This compares to SCSI,
which is currently running at 320 MBps, an increase from previous SCSI speeds
of 160 MBps and 80 MBps. Although Ultra 160 SCSI and Ultra 320 SCSI have
high bandwidth, as parallel protocols they lack several important features that
are useful for networking.
Another benefit of Fibre Channel is that it is bi-directional, which means that it
can send and receive data at the same time. This feature effectively doubles the

237
The SAN Book III – October 22, 2001

throughput of Fibre Channel. Although Fibre Channel is still superior for


networking, SCSI is proving to be a formidable competitor for the disk drive
interface.
Because Fibre Channel ➤ Greater Connectivity. Because Fibre Channel supports multiple topologies,
supports multiple such as point-to-point, arbitrated loop, and switched, it can scale up to meet the
topologies, it can scale
up to meet the
requirements of any particular application. For example, in arbitrated loop
requirements of any mode, Fibre Channel can connect up to 126 nodes — a theoretical 16 million
particular application. nodes in switched mode — compared with SCSI, which connects only 16.
➤ Centralized Management. Switched architectures can be managed through one
central point; as such, bandwidth can be allocated according to demand.
Centralized management is especially important for applications involving e-
commerce and transaction processing.
➤ Low Overhead (Low Latency). In addition, the Fibre Channel protocol has very
little transmission overhead, compared with SCSI, IDE, and TCP/IP (Ethernet).
This reduces latency and makes the hardware usage highly efficient, delivering
greater price/performance.
➤ Increased Reliability. Fibre Channel guarantees the delivery of data, adding a
high degree of reliability, which is critical for the intense data integrity
requirements of storage applications. This reliability makes Fibre Channel a key
differentiator for use in SANs, when compared with traditional LAN (Ethernet)
or WAN (ATM) technologies.
➤ Distance and Cabling. Fibre Channel cables inherently go farther (up to 10 km)
and with greater ease (cables are less bulky and easier to handle) than SCSI.
There are also fewer impedance issues, which improves data integrity (such as
using one wire rather than many intertwined). The maximum cable length is an
important criterion for campus-level SANs.
Does this mean that SCSI is dead? No! Or at least not for a while. In our
experience, technological shifts of this magnitude occur over long periods of time
and, in many cases, both technologies co-exist for prolonged periods. We would
refer to this migration as evolutionary, not revolutionary. We expect both
connectivity solutions to co-exist for at least another five to ten years, albeit at a
diminishing rate.

238
The SAN Book III – October 22, 2001

Fibre Channel Versus Ethernet

➤ Fibre Channel has won out over SCSI for large-scale SANs. The next
debate is whether Ethernet can beat Fibre Channel for future growth.

➤ We explain Fibre Channel and the basics of protocols in this section


so investors can better evaluate the IP storage debate.

What’s the Big Deal?


So if Fibre Channel is so good, why isn’t everyone doing it? How could the largest
networking companies on the planet miss such an important networking technology
shift that we expect will drive growth for the next decade? Simple — they didn’t.
In fact, 3Com was one of the first networking companies to sign up for the first
Fibre Channel standards body meeting with the Internet Engineering Task Force
(IETF) in 1996.
Fibre Channel was expected to take off in 1997. After allocating resources for
several years, 3Com dropped its Fibre Channel initiative, sending a signal to other
networking companies that there was little market potential. Many other large
networking companies took 3Com’s lead...”Yeah I had heard of Fibre Channel, but
3Com spent a couple of years doing diligence and dropped out. So we thought
nothing would happen and that storage would have to wait for Ethernet to increase
its bandwidth to provide a networked infrastructure.” Other traditional networking
companies such as Cisco were also busy with other initiatives such as voice-over-IP.
The other important factor was NAS, which was just beginning to gain traction.
NAS already used IP so there seemed to be no need for a new storage networking
protocol.
Two years later, in 1999, Fibre Channel began to blossom. But since this seemingly
niche market (including HBAs) was only projected to reach $3 billion in 2003 (by
Dataquest who has since increased its expectations to $7 billion in 2003 and $17
billion in 2005), the established networking companies looked to NAS and 10
Gigabit Ethernet as the answers that would slam the fledgling Fibre Channel market
back to whence it came. The problem was that Fibre Channel technology began to
embed itself into other storage components, such as disk drives, RAID and servers
and backplane technologies. The other, and larger problem, was that 10 Gigabit
Ethernet did not solve the performance issues Ethernet presented: In other words, it
is not simply about speeds and feeds.
So what do you do if you are a thousand-pound gorilla and you are late to the game?
You partner with the leaders, buy your way in, change the rules, or all of the above.
Aside from the hope that 10 Gigabit Ethernet would throw enough bandwidth at the
problem to make the discussion moot, many incumbents banded together to form an
IP alliance and various IP protocol proposals began to flood the IETF.
In order to frame the debate (like the pun?), let’s first review what it is, exactly, that
we are talking about. In other words, let’s review what a “protocol” is. Then let’s

239
The SAN Book III – October 22, 2001

look at what today’s TCP/IP Ethernet is and compare it to today’s Fibre Channel.
Then let’s see whether, and how, Ethernet might work out in the house of storage.

Understanding Networking Protocols and Architectures


Protocol discussions can be confusing since the same term can have different
meanings. For example, the term “Ethernet” can refer to a protocol, a protocol
stack, or an architecture.
What Is a Protocol?
A protocol is a set of A protocol is a set of rules that multiple devices adhere to for sending and receiving
rules that multiple data to and from each other. These rules can describe specifications such as how
devices adhere to for
sending and receiving
data should be addressed, how error checking should be accomplished, or how data
data to and from each should be segmented for transport. Many protocols are open standards that vendors
other. follow so they can communicate with devices from other vendors. For example, in
order for a computer to send data to a disk drive, it can use the SCSI, IDE, Fibre
Channel or ATA protocol.
What Is a Protocol Stack?
A protocol stack is a combination of protocols that work together. For example, the
TCP/IP protocol stack includes the TCP, IP, and Ethernet protocols. We will use the
term Ethernet to refer to the TCP/IP stack. Note: The TCP/IP stack technically
includes only TCP and IP, but since Ethernet is often used with this stack in the real
world, the industry frequently refers to this combination as Ethernet or IP.

What Is an Architecture?
Architecture refers to the Architecture refers to the hardware and software used to implement the protocol.
hardware and software For example, a network interface card (NIC) is part of the Ethernet architecture.
used to implement the
protocol.
Using Layers to Describe Protocols
What Are Layers?
Layer diagrams are often used to describe protocols. In order to send data across a
network, different layers of instructions must be added. As the data packet goes
through the different layers, it grows larger as more information is added.
Figure 238 below illustrates how data is networked using the OSI model which we
describe in further detail in the next section. Each discrete set of instructions is
represented by a separate layer in sequence. Note: Although OSI divides the
instructions into seven categories, any number of categories can be used.

240
The SAN Book III – October 22, 2001

Figure 238. OSI Reference Model

Data Networking Info Networking Info Data

Data Sent Data Received

Source: Salomon Smith Barney


The final data packet, complete with all instructions (such as the address) and the
original data (also referred to as the payload), becomes what is known as a frame in
Fibre Channel parlance and a packet in Ethernet. The frame (assuming it is Fibre
Channel) is then finally sent to the other node (which can be a computer or device).
As you can imagine, this process can create quite a bit of overhead.
Once the information crosses the wire and arrives at the other node (a node can be a
computer or device), the frame must go through the exact same steps but in reverse
order. At each layer, information is stripped away, the frame decreases in size, until
the final application gets the raw data that was originally sent.
OSI Model
The OSI is a generic The International Standards Organization (ISO), which focuses on standardization in
model which categorizes technology, created the Open System Interconnection (OSI) Reference Model (see
the instructions and
processing involved in
Figure 239 below). The OSI is a generic model which categorizes the instructions
networking into seven and processing involved in networking into seven distinct, sequential layers. To
distinct, sequential understand which instructions different protocols describe, it is useful to compare
layers.
them using the OSI model.
Figure 239 below illustrates how Ethernet and Fibre Channel compare using the OSI
model. Ethernet provides guidelines for layers one through four of the OSI model.
Fibre Channel provides guidelines for layers one through five. Other higher-level
protocols must be used to fill in the rest. Note: The relationship to the OSI model is
not perfectly matched. In Ethernet, the TCP protocol includes some elements of
both the transport and networking layers. In Fibre Channel, FC-2 includes
elements of the networking and transport layers and FC-4 includes elements of the
transport and session layer.

241
The SAN Book III – October 22, 2001

Figure 239. OSI Versus Ethernet Versus Fibre Channel

OSI Fibre
Model Channel Ethernet
L7: Application
Data L6: Presentation

L5: Session
FC-4
L4: Transport TCP

L3: Network IP

L2: Datalink FC-3 Ethernet

FC-2
FC-1

L1: Physical FC-0

Source: Salomon Smith Barney

Figure 240 below describes the type of processing that occurs on each layer.
Note: We have seen multiple ways of mapping Fibre Channel, Ethernet, and other
storage protocols to the OSI model from various respected sources. Since the OSI is
an idealized model which no one is required to adhere to, protocols do not always
match it exactly. For example, protocols can add new rules which are not included
in the OSI model, or they can group the layers differently. Since protocol mapping
is more of an art than a science, this represents our best approximation.
Figure 240. ISO’s OSI Model
Layer What it Does Analogy
L7) Application Translates the message into a form that Draft a letter.
the receiving computer will understand.
For example, text into ASCII.
L6) Presentation Translates, compresses and encrypts the Print out actual message on
message. stationery paper.
L5) Session Details about the communications Will it be sent first or second class?
session.
L4) Transport Segments the data. Breaking up the parcel into
separate packages.
L3) Network Adds addressing and routing information. Write down address and best way
to get there.
L2) Datalink Prepares data for particular method of If sending first class, put inside
physical medium. First class envelope.
L1) Physical Transfers from computer onto medium Send it out of the building via truck.
(i.e., copper, optical).
Source: Salomon Smith Barney

242
The SAN Book III – October 22, 2001

Sending data from host Sending data from host to target can be analogous to one person sending a letter to
to target can be another. Data begins to be processed from top to bottom, from layer seven to layer
analogous to one person
sending a letter to
one. The first three layers deal with the presentation of the data before it is broken
another. up into smaller packets of data. In layer seven, the computer transforms the raw data
into a form that can be sent, similar to writing your thoughts down in a draft letter
that can be read by someone else. In layer six, it formats it even further, similar to
transforming your draft into a nicely written letter that can be properly sent. In layer
five, it describes features of the communication session (i.e., security), similar to
deciding if your letter should be sent first or second class.
The next four layers deal with breaking up the data into smaller packets and
preparing these packets to be reassembled correctly upon arrival. In layer four, the
data is segmented into smaller pieces, similar to breaking down a parcel into
separate boxes. In layer three, the computer adds on addressing and routing
information, similar to writing down the address of the person you are sending your
letter to on the envelope, as well as the return address so the recipient knows who it
is coming from. In layer two, it prepares the data to be used properly depending on
what kind of physical medium it is travelling on, similar to putting the letter inside a
first class envelope so it can be sent correctly. In layer one, the data finally gets sent
onto the wire, similar to the post office putting the letter into the truck and out on the
road.

Fibre Channel Basics


Fibre Channel describes layers one through five of the OSI model (see Figure 241
below). Note: FC-2 has elements of OSI layer four so it is not a perfect one-on-one
correlation between the OSI and the Fibre Channel layers as depicted below. FC-4
occupies the network, transport and part of the session layer.
Figure 241. Fibre Channel Versus OSI Model

OSI Fibre
Model Channel
L7: Application
Data L6: Presentation

L5: Session
FC-4
L4: Transport

L3: Network

L2: Datalink FC-3

FC-2
FC-1

L1: Physical FC-0

Source: Salomon Smith Barney

243
The SAN Book III – October 22, 2001

Figure 242 below describes the functions of each layer.


Figure 242. Fibre Channel Five-Layer Model
Layer Description
FC-4) Upper-Layer Protocol Interface SCSI is the most commonly used
FC-3) Common services Advanced features
FC-2) Data Delivery Framing and flow control
FC-1) Byte Encoding Encode and decodes eight bits to ten
bits so the transmission will be more
reliable.
FC-0) Physical layer Optical or copper wire may be used.
Source: Salomon Smith Barney

The Fibre Channel layers carry out the following functions:


➤ FC-4: Upper-Layer Protocol Interface. The first step is to generate the data.
SCSI generates a command or data to send across the network. SCSI includes
storage-related commands such as “write to disk.”
➤ FC-3:Common Services. This layer defines advanced features such as striping
(to transmit one data unit across multiple links) and multicast (to transmit a
single transmission to multiple ports).
➤ FC-2: Data Delivery. This layer contains basic rules needed for sending data
across a network. This includes: 1) how to divide the data into smaller frames,
2) how much data should be sent at one time before sending more (i.e., flow
control), and 3) where the frame should go. It also includes Classes of Service
which define different implementations of Fibre Channel that can be selected
depending on the application (for more detail, see the section titled “Guaranteed
Delivery.”
➤ FC-1: Byte Encoding: This layer takes the frames from step FC-2 and
transforms it so it can be sent across a wire with less chance of error.
➤ FC-0: Physical Interface: This layer is part of the Fibre Channel that says
which type of physical media can be used in the network. For Fibre Channel,
either optical of copper cable may be used.

Backtracking Through a Little Fibre Channel History


In the mid-1990s, some Backing up in time for a moment, Fibre Channel actually came to the market in the
companies began to late eighties as a data communications networking technology; in other words,
realize the benefits of
Fibre Channel.
instead of using Ethernet one would use Fibre Channel to connect servers and PCs.
In fact, Ancor Communications (now owned by QLogic), one of the current leaders
in Fibre Channel networking, was founded in the late eighties as a LAN company.
The benefit of Fibre Channel is that it can transmit larger packets of data effectively
(referred to as “block level” data transfers), while Ethernet is optimized for small
packets. In the end, Ethernet won because of its backing by the government and its
broad-based support from multiple organizations. Note: There was no real need at
the time to send block level data over the LAN. That was being handled behind the
server by SCSI, which at the time did not have a demand to be networked.

244
The SAN Book III – October 22, 2001

In the mid-1990s, when SCSI began to look as though it was running out of gas,
companies such as EMC and Brocade began to realize the benefits that Fibre
Channel could bring by networking storage.

Ethernet Feature Comparison


The Ethernet protocol stack (TCP, IP, and Ethernet protocols) describes layers one
through four of the OSI model (see Figure 243 below). It is the most widely-used
networking protocol stack for LANs. Note: We will use the term “Ethernet” to refer
to the protocol stack and the term “Ethernet protocol” to refer to the protocol itself.
Figure 243. Ethernet Versus OSI Model

OSI
Model Ethernet
L7: Application
Data L6: Presentation

L5: Session

L4: Transport TCP

L3: Network IP

L2: Datalink Ethernet

L1: Physical

Source: Salomon Smith Barney

Figure 244 below briefly outlines the main function of each of these protocols in
Ethernet.
Figure 244. Ethernet — Protocol Functions
Layer Description
TCP Ensures reliable transport
IP Addressing
Ethernet Formatting data for physical transport
protocol
Source: Salomon Smith Barney

TCP/IP was originally design by the U.S. Department of Defense and largely used in
governmental agencies and universities. It was the use of TCP/IP to develop the
Internet, however, that gave it the ultimate upper hand over competitors such as
Novell’s SPX/IPX and IBM’s SNA protocols.

Ethernet Features
Features of Ethernet include: 1) different framing architecture, 2) longer instruction
sets, 3) transmitting individual packets on different routes versus sequences of
packets on the same route, and 4) different delivery guarantees.

245
The SAN Book III – October 22, 2001

1. Different Framing Architecture


Gigabit Ethernet has packet sizes that range between 512 bytes and 1,522 bytes (see
Figure 245 below). Note: Each segment does not correspond to a separate OSI
layer. The figure is not drawn to scale as the data payload can be as large as 1500
bytes.
Figure 245. Ethernet Packets

Destination Source Type/ Frame


Data
Address Address Length Check Seq.

512 to 1,522 bytes

Source: Salomon Smith Barney

If the data to be If the data to be transmitted is smaller in size, the packet is extended to 512 bytes
transmitted is small in through a technique called “carrier extension,” which adds extension symbols for
size, the packet is
extended to 512 bytes
collision detection purposes. The extra bytes are overhead and reduce overall
through a technique efficiency. Note: Ethernet jumbo frames allow for packets as large as 9000 bytes
called “carrier but are not commonly used or supported by all existing infrastructure.
extension.”
In Figure 246 below, the light grey areas have all of the relevant information that is
needed to send the data payload. However, the entire length of this data string is
less that 512 bytes. Hence extension symbols are added at the end to make sure the
packet is exactly 512 bytes long.
Figure 246. Ethernet Packet Adds Extension Symbols for Small Data Payloads

Destination Source Type/ Frame Extension


Data
Address Address Length Check Seq. Symbols

64 bytes

512 bytes

Source: Salomon Smith Barney

Fibre Channel is Fibre Channel, on the other hand, has frame sizes that range between 36 and 2,148
effective for high- bytes (that’s approximately 2 kilobytes). The amount of overhead, however,
volume, small frame
transmission, and also
remains constant regardless of the size of the data to be transmitted. This makes
has sequencing Fibre Channel highly efficient for sending high volumes of smaller frames.
capabilities for transfers
of large blocks of data. In addition, Fibre Channel frames can be linked together in a sequence, which makes
the protocol useful for large bulk transfers of data. Note: Gigabit Ethernet also has
frame bursting, which allows for the transmission of a series of frames at one time.
Figure 247 below illustrates a Fibre Channel frame. Unlike an Ethernet packet, the
Fibre Channel frame does not waste transmission cycles with extension symbols.
Both Ethernet and Fibre Channel frames contain the same type of information.

246
The SAN Book III – October 22, 2001

Note: Although the sections of the Fibre Channel frame in Figure 247 below are
labeled differently from those in the Ethernet packet above, both have the same type
of information, not including the extension symbols. The header includes the source
address, destination address and type; similar to Ethernet. The figure is not drawn
to scale as the data payload can be as large as 2112 bytes.
Figure 247. Fibre Channel Frames

Cyclic
Start of End of
Header Data Redundancy
Frame Frame
Check

36 to 2,148 bytes

Source: Salomon Smith Barney

2. Longer Instructions
Ethernet works well for small bits of information, but generates too much overhead
to effectively transmit larger blocks of data. It has a more cumbersome instruction
set (over 5000 instructions per stack versus a few hundred in Fibre Channel) which
requires the utilization of more CPU cycles than Fibre Channel to process a packet.
Additionally TCP/IP is typically processed using software. Fibre Channel’s
instructions are processed more quickly by using hardware (the HBA). Several
companies, however, have announced plans to develop TCP accelerators, which
process sections of the TCP/IP stack on hardware.
Also note that Gigabit Ethernet, although it raised its data transfer rate 10x by
moving to 1 Gbps, it only increased its throughput by approximately 3x as a result of
its high overhead (one could argue the increase is 2x or 4x, depending on the method
of measurement).
While TCP/IP does the 3. Sends Packets over Different Routes
job for file level data on
TCP/IP often routes each packet differently, giving each individual packet of a given
the LAN, it does not
currently meet the data transmission a different route to arrive at the same destination as the others;
performance whereas Fibre Channel logically groups frames and sends several at a time in
requirements of block sequence, ensuring they arrive in order. The result is that TCP/IP often drops
level storage
transmissions.
packets of data when the network becomes congested. The packets must then be
retransmitted using more bandwidth.

4. Regulated Flow Control


Flow control guidelines specify how many packets to send at a time. In TCP/IP,
packets are sent without an indication if the receiver is ready for it. In Fibre
Channel, the sender asks the receiver how many frames it has capacity for. Once it
gets an answer, it will send the number of frames allowed, in sequence.

5. Guaranteed Delivery
TCP/IP always offers guaranteed delivery, a feature in which the receiver will send
an acknowledgement to the sender after each packet. If the sender does not get the
acknowledgement within a specified period of time, it will retransmit the packet.

247
The SAN Book III – October 22, 2001

Hence, when two stations are transmitting data on the same line at the same time, no
acknowledgement will be received by either. Each station will then wait a random
amount of time before transmitting again. In heavy traffic conditions, after ten
unsuccessful attempts, the delay time increases. After 16 unsuccessful attempts, the
station gets an error and no more tries are made. Although this reduces congestion
on a busy network, it will also slow down data transmission.
Fibre Channel offers six Fibre Channel offers six different classes of service that can change the delivery
different classes of service guarantee characteristics depending on the type of environment the SAN is in. Class
that can change the
delivery guarantee
1, a circuit-switched connection, is a dedicated, uninterruptible link, analogous to a
characteristics depending telephone connection. The benefit of Class 1 service is that transmissions are
on the type of environment sustained and guaranteed, which is useful for high-volume, time-critical
the SAN is in.
transmissions, such as between two large supercomputers. Class 2 service
guarantees delivery with receipt confirmation, but no dedicated connection is
established. Class 3 is similar to Class 2 but without a guarantee of delivery so there
is no wait for confirmation. Note: In Class 3, the most common implementation of
Fibre Channel, there is no guaranteed delivery, which means Fibre Channel
immediately resends packets upon receiving an error.

Example of Ethernet Versus Fibre Channel


Okay, let’s look at the different routing, flow control, and guaranteed delivery
features in a way that is easier to understand. Let’s say you have to send a group of
children to a museum. Ethernet would have each child find his/her own way to the
museum. Along the way, one might get lost, delayed, or destroyed (remember, this
is really data we are talking about). If everyone arrived, Ethernet would have to
reorganize them back in alphabetical order and then wait for the museum to open.
Fibre Channel, on the other hand, would first put the kids in alphabetical order and
have them hold hands so they would not go out of order. It would call the museum
beforehand to tell them when it was arriving so the museum would know when to
open and how many people to expect. Then each kid would get on the bus in an
orderly manner. The bus would arrive at the museum exactly on time. The children
would arrive in the exact same order as when they started, and the transaction would
be completed.

New Ethernet-Based Protocols Are More Suited for Storage


Although Ethernet has several features that make it more suitable for networking
small messages versus entire blocks, new protocols are being proposed which
modify Ethernet to make it more suitable for storage. In the next section we discuss
three: 1) iSCSI, 2) iFCP, and 3) FCIP.

248
The SAN Book III – October 22, 2001

Potential Future IP Storage Iterations


➤ There are several IP storage standards being reviewed by the IETF.
Three of the more popular ones are iSCSI, iFCP, and FCIP.

➤ We believe IP storage will initially occupy the periphery of the market


before it can ramp significantly and potentially compete head-to-
head with Fibre Channel in the enterprise.

Introduction
The Internet Engineering Task Force (IETF), a standards body for Internet
architectures, is currently considering several standards for IP storage. We use the
term IP storage to describe protocols that are based on IP over Ethernet networks
versus Fibre Channel. Although network-attached storage (NAS) appliances also
work over Ethernet, we do not include NAS per se in our definition of IP storage.
Three of the more Three of the more prevalent IP storage protocols are iSCSI, iFCP, and FCIP.
prevalent IP storage Understanding the differences between these proposals can be helpful in comparing
protocols are iSCSI,
iFCP, and FCIP.
the unique benefits they have to offer.
In this section, we explain IP storage and what stages we believe this technology
will go through in the future.

Why Use IP Storage?


If Fibre Channel is currently working well, why use IP/Ethernet? According to IP
storage proponents, three of the main reasons are:
➤ Performance: IP storage could eventually be both cheaper and faster since more
money and resources are being spent on advancing Ethernet than Fibre Channel
due to a larger installed base. Cisco and IBM, two influential companies, are
currently supporting iSCSI.
➤ Cost Savings: Eliminates the need to have two networks. One super network
interface card (sNIC) can be used to access both Ethernet networks and IP
SANs. Two cards are needed, a NIC and a host bus adapter (HBA), to access
Ethernet and Fibre Channel networks, respectively.
➤ Manageability: Eliminate the need to have network administrators learn
multiple ways to control data networking and storage networking.

Three Versions of IP Storage: iSCSI, iFCP, and FCIP


There are three IP storage protocols currently being considered by the IETF: 1)
iSCSI, 2) iFCP, and 3) FCIP. ISCSI and iFCP are native protocols that transport
SCSI over Ethernet while FCIP is a tunneling protocol that encapsulates Fibre
Channel in Ethernet.
Fibre Channel
As a quick review, Fibre Channel takes SCSI data and commands and adds new
information onto these items so they can travel across a networked serial interface,
similar to data across an Ethernet network.

249
The SAN Book III – October 22, 2001

iSCSI
iSCSI sends SCSI iSCSI sends SCSI commands as data, (as does Fibre Channel), but to transport the
commands as data, (as data, it uses Ethernet layers instead of the Fibre Channel layers. Additionally, it uses
does Fibre Channel), but
to transport the data, it
a new protocol called iSCSI to replace the FC-four layer that Fibre Channel uses.
uses Ethernet layers This protocol can be used on IP SANs only.
instead of the Fibre
Channel layers. Figure 248 below compares Ethernet, iSCSI, iFCP, and Fibre Channel. Note:
Protocol mapping is more of an art than a science and interpretations may vary.
Figure 248. Ethernet, iSCSI, iFCP, Fibre Channel

Fibre
Channel Ethernet iSCSI iFCP

FC-4 iSCSI FC-4 (SCSI)


TCP TCP TCP
IP IP IP

FC-3 Ethernet Ethernet Ethernet

FC-2
FC-1

FC-0

Source: Salomon Smith Barney

Figure 249 below illustrates how newer protocols take elements of older ones.
ISCSI takes elements of both SCSI and IP, similar to the way in which Fibre
Channel took SCSI and morphed it into Fibre Channel by adding networking
characteristics.
Figure 249. Newer Protocols Use Elements of Older Ones

Older Newer

SCSI FC

Ethernet iSCSI
Source: Salomon Smith Barney

iFCP
The difference between Similar to Fibre Channel and iSCSI, iFCP sends SCSI commands as data but to
iFCP and iSCSI is that transport the data, it uses Ethernet layers instead of Fibre Channel layers. The
iFCP keeps the FC-4
layer instead of replacing
difference between iFCP and iSCSI is that iFCP keeps the FC-4 layer instead of
it with iSCSI. replacing it with iSCSI.

250
The SAN Book III – October 22, 2001

The FC-Four Layer Enables Backward Compatibility


By using an FC-four layer instead of an iSCSI layer, iFCP can be used with current
software applications that manage Fibre Channel networks. This protocol can be
used on both IP and Fibre Channel–based SANs. As you can see in Figure 250
below, different protocols have different degrees of backward compatibility. We
believe this is an important feature that IT managers will consider when deciding
which technology they will use in the future.
Figure 250. Technologies Have Different Degrees of Backward Compatibility

Fibre iSCSI iFCP FCIP


Channel
Fibre Channel SANs yes no yes yes
IP SANs no yes yes yes
Works with Fibre yes no yes yes
Channel software
Source: Salomon Smith Barney

FCIP
The method of taking a complete frame in a different protocol and wrapping it in yet
another protocol is known as tunneling. FCIP is an open, standard way of tunneling
Fibre Channel through IP. Proprietary methods of tunneling also exist.
FCIP takes a Fibre FCIP is a combination of Ethernet and Fibre Channel. It takes a Fibre Channel
Channel frame of frame of information and then adds Ethernet layers to it. This results in more data
information and then
adds Ethernet layers to it.
needed to be transmitted for a given file (see Figure 251 below) than in either
protocol by itself because after adding all of the Fibre Channel layers, it then must
add several IP layers of instructions.

251
The SAN Book III – October 22, 2001

Figure 251. FCIP Combines Two Protocols

FC FCIP Ethernet

FC-4 FC-4
FC-3 FC-3
FC-2 FC-2

FC-1 TCP TCP

IP IP

FC-0 Ethernet Ethernet

Source: Salomon Smith Barney

We See IP Storage Becoming a Complement to Fibre


Channel
Over the next few years, Over the next few years, we believe IP storage will be more of a complement to
we believe IP storage will Fibre Channel than a competitor. We believe it will emerge in the low end of the
be more of a complement
to Fibre Channel than a
market because by the time it becomes a viable alternative to Fibre Channel, most
competitor. enterprise customers who need the technology now and can afford it will have
already made significant investments in Fibre Channel. In addition, the proposed
iFCP and FCIP protocols could be valuable complements to Fibre Channel by
allowing geographically dispersed Fibre Channel SANs to communicate by
“hitching a ride” on an existing Ethernet backbone.
Although TCP/IP has been around for a long time, we believe IP storage will take
approximately two to five years to equal the performance of Fibre Channel. We
believe the following things need to happen before IP storage can become a
mainstream technology:
➤ IETF must approve a standard. This standards board is currently in the
process of reviewing various proposals, each of which has many different
versions.
➤ Vendors must modify products. Vendors such as Brocade, Compaq, EMC,
IBM, Sun, and Veritas must modify their products before they can work with the
iSCSI standard.
➤ Early adopters must test. New technologies take time to debug. Several
upgrade cycles could occur before IP storage is mainstream.
➤ Complementary technologies must be developed. In order for IP Storage to
equal the performance of Fibre Channel, new technologies must be developed to
allow for faster processing and transmission. We believe two of the more
important technologies are 1) increased bandwidth over Ethernet, and 2)
hardware acceleration of IP processing. Fibre Channel currently runs at 2 Gbps
and Ethernet at 1 Gbps, so Ethernet technology must surpass Fibre Channel to
equate its performance. Additionally, hardware accelerators to process the
longer IP instructions will make processing the data quicker. Currently IP is

252
The SAN Book III – October 22, 2001

processed in software compared to Fibre Channel’s instruction set, which is


processed in hardware.

IP Storage Is a Great Choice at the Department Level


We believe IP storage can be a great choice at the department and workgroup level,
much as NAS is today. It means companies who have small budgets and limited IT
resources will not have to buy additional HBAs on top of their NICs, thereby saving
money. Also, they will not have to retrain IT managers who already understand
Ethernet, thereby reducing complexity.
Greater resources are In time, IP Storage’s performance could surpass that of Fibre Channel. Greater
being devoted to resources are being devoted to Ethernet’s research and development by large
Ethernet’s research and
development by large
traditional networking companies, such as Cisco.
traditional networking
However, the enterprise customers who are the main customers of Fibre Channel
companies, such as
Cisco. technology today could be reluctant to completely dispose of their Fibre Channel
investments in order to use a newer technology. Until IP Storage adoption is
complete, more money will continue to be invested into Fibre Channel, making it
more expensive to switch to IP Storage with each passing day. IT managers could
be reluctant to switch to IP Storage until its price/performance is a magnitude greater
than Fibre Channel, and not simply equal to it, and we believe this could take many
years.
We believe new technologies follow a cycle similar to Gartner’s Technology
“Hype” Cycle (Figure 252). The chart illustrates the amount of attention a
technology attracts over time.

253
The SAN Book III – October 22, 2001

Figure 252. New Technology “Hype” Cycle and Protocol Positioning

Peak of inflated
expectations
Plateau of
productivity

Attention Slope of
enlightenment
IP Storage

Trough of
Fibre
disillusionment
Channel
Trigger Fails to establish
compelling end-user
value
InfiniBand

Time
Source: Salomon Smith Barney for protocol positioning. Dataquest for Technology “Hype” Cycle.

In the early days of Fibre Channel people believed it was going to do everything
including the file sharing. We believe IP storage will go through a similar evolution.
In its infancy, IP storage is currently being hailed as the all-encompassing
replacement to Fibre Channel, and is rapidly attracting attention. We believe it will
eventually occupy a solid segment of the market and co-exist with Fibre Channel.
According to industry pundit Mike Gluck, “IP storage will happen slower than the
‘hype’ but faster than Fibre Channel evolved.”

254
The SAN Book III – October 22, 2001

The Future Band: InfiniBand


➤ We believe InfiniBand is on its way to becoming the next generation
I/O. InfiniBand solves a painful computing bottleneck while
increasing performance and scalability.

➤ We expect to first see InfiniBand as an HBA solution and then


migrate to a chip based solution that (effectively) externalizes the I/O.

InfiniBand: Bringing Input/Output Outside


InfiniBand, previously named System I/O (referred to as SIO), is the result of the
merging of two competing, developmental I/O (Input/Output) architectures:
1 Next Generation I/O (NGIO), which was architected by Intel and supported by,
for example, Dell, Hitachi, NEC, Siemens, and Sun; and
2 Future I/O (FIO), which was being backed by companies including 3Com, Cisco,
Compaq, HP, IBM, and Adaptec.
The goal of InfiniBand is to solve today’s server I/O bottleneck with a new
computer architecture/connection that utilizes an InfiniBand switching fabric in
place of the current PCI Bus structure.
Below in Figure 253 we highlight the major InfiniBand supporters. Membership of
the InfiniBand Trade Association currently includes 230 companies across diverse
product fields including server, networking, storage, software, and silicon providers.
Figure 253. Premier InfiniBand Supporters

Steering Committee Companies

Sponsoring Member Companies

Source: InfiniBand Trade Association

The PCI and PCI-X Bus architectures create an inefficient bottleneck as data gets
congested as it passes through the bus. InfiniBand overcomes this by implementing
a switched fabric chip architecture to direct and control the flow of data; thereby,
eliminating the PCI bottleneck. InfiniBand proposes to offer greater connectivity

255
The SAN Book III – October 22, 2001

with increased performance by enabling data to flow into and out of devices more
efficiently.

Evolution of Interior I/O


The only constant is With the exponential growth of bandwidth outside the server (Gigabit Ethernet plus
change. advanced storage networks) and the increasing speeds of microprocessors and
memory architectures, one thing is abundantly clear: Current I/O interconnects need
an overhaul. While SAN solutions are certainly interesting out-of-the-box solutions,
evolution inside the box is also unfolding. The two current development efforts
center around PCI-X, which is an evolutionary bus structure, to PCI, and InfiniBand,
an entirely new switched fabric I/O architecture.
PCI-X Is Most Likely Next in Line
While InfiniBand and Gigabit Ethernet may get more air time, PCI-X will most
likely be the next technology milestone to be reached. When Intel invented the PCI
bus in 1991, it was because the ISA and EISA bus structures were reaching their
bandwidth limits. Now, almost a decade later, as microprocessor frequencies have
risen from under 100 MHz to over 1 GHz, the limits of the PCI bus have been
reached. Given that system OEMs such as Dell, Compaq, IBM, and Hewlett-
Packard will likely be launching PCI-X systems prior to or in 2002, it’s important
for the HBA suppliers to grow in concert.
Doubling the Bandwidth of PCI
PCI-X improves upon PCI PCI is a ubiquitous bus interface used throughout computing (desktops,
in both speed and workstations, servers) and is even moving into communications equipment such as
number of
expansion slots.
routers and switches. Graphics cards, modems, Ethernet NICs, SCSI/Fibre Channel
Host Bus Adapters are all common residents of PCI slots today. The latest
generation of PCI (version 2.2) is a 64-bit bus capable of 66 MHz speeds; peak
bandwidth tops out at 512 MBps. PCI-X is a 64-bit, 133 MHz bus capable of 1
Gbps performance, double that of PCI 2.2.
PCI-X not only runs at higher frequencies, but also supports more slots. PCI 2.2
systems, at 66 MHz, can support only one or two slots (with additional slots, the bus
speed drops down to 33 MHz). PCI-X improves upon this limitation, offering
designers flexibility in choosing between one slot at 133 MHz, two at 100 MHz, or
three or more slots at 66 MHz.
Figure 254. PCI-X Delivers Higher Bandwidth than PCI
Bus width Bus Frequency Peak Bandwidth
Specification
(bits) (MHz) (MBytes/sec)
PCI 2.2 32 33 133
PCI 2.2 64 33 266
PCI 2.2 64 66 533
PCI-X 1.0 64 133 1,066

Source: Salomon Smith Barney

256
The SAN Book III – October 22, 2001

Additional Efficiency Features


PCI-X improves The designers of PCI-X also developed bus utilization features, which improve
bandwidth over PCI, traffic flow and efficiency of data transfers on the bus. Probably the most significant
even at the same
frequency.
feature is the addition of “split transactions,” which solves a major drawback of PCI
called “delayed transactions,” which reduce system bandwidth. Other features
include more stringent rules for inserting wait states, which also reduce bandwidth,
and allowable disconnects. These and other enhancements greatly improve the bus
efficiency so that a PCI-X system will have greater bandwidth than a PCI system,
even if it runs at the same bus frequency.
PCI-X Support Is on the Way
The Fibre Channel HBA PCI-X will be fully backwards-compatible with PCI, meaning that PCI-X devices
makers that ship PCI-X need to work in PCI slots and PCI devices need to work in PCI-X slots. So the
products first will have a
slight advantage.
challenge for designers is not only to implement the new features of PCI-X, but also
to remain 100% backwards-compatible — not a trivial undertaking. However, as
server OEMs increasingly look for ways to improve system bandwidth, PCI-X
support will become increasingly important.
Figure 255. PCI-X Implementation in Next-Generation Servers

Source: Agilent

3GIO and HyperTransport are two other I/O technologies that are being discussed as
replacements for PCI, but are beyond the scope of this report.

To InfiniBand and Beyond


While PCI-X will satisfy I/O bandwidth for the next couple of years, a longer-term
vision reveals that a quantum leap forward for I/O architectures will likely be
necessary to satisfy continually increasing data throughput requirements. Andreas
Bechtolsheim, vice president of Engineering at Cisco, commented that “PCI will just
barely be able to handle 1 Gb Ethernet… PCI-X will be just barely able to handle 10
Gb Ethernet. To scale much beyond that dictates a shift to a new sort of I/O
protocol.” To this end, Intel and a consortium of server and I/O components
vendors are paving the road for InfiniBand, a new switched I/O architecture that
may eventually bring the I/O interconnect outside the server (meaning, no more
internal slots).

257
The SAN Book III – October 22, 2001

What Is InfiniBand?
InfiniBand proposes to bring these I/O connections “outside” the servers, so
computing power can be upgraded or added in a simple, efficient manner. In
addition, I/O expansion takes place across a switched fabric environment,
eliminating slot limitations of PCI and PCI-X.
IDC projects servers with InfiniBand capability will grow to 6,720,296 in 2005 from
3,806 units in 2001, representing a 346.1% CAGR (Figure 256 below). Although
we believe this is a bit optimistic, it drives home the point that InfiniBand is coming
and is going to be a big market opportunity.
InfiniBand is a complete Figure 256. InfiniBand-Enabled Server Shipments
new architecture; the
transition won’t be easy
or quick. Units
7,000,000

6,000,000 346.1% CAGR


5,000,000

4,000,000

3,000,000

2,000,000

1,000,000

0
2001 2002 2003 2004 2005
Source: IDC (May, 2001)

But what exactly is it? Is it a new bus or a protocol? Does it reside in the server or
outside? InfiniBand is an architecture based on a switched-fabric of serial I/O
channels; InfiniBand is so broad-reaching that it redefines existing I/O systems,
connections, cables, physical layer and high-level software protocols. Various I/O
subsystems such as Ethernet, Fibre Channel, SCSI, or even inter-processor
communication will be able to link to an InifinBand fabric. But the transition won’t
be easy or quick nor is it guaranteed to happen. Systems supporting InfiniBand will
likely first be introduced in 2002, even under the most optimistic estimates.
Think Outside the Box, Literally
InfiniBand moves the I/O A significant transition to be brought about by InfiniBand is the decoupling of the
subsystem out of the I/O subsystem from the server’s microprocessor/memory complex. Today’s server
server.
I/O subsystem consists of shared PCI bus, linked directly to the system memory and
CPU. InfiniBand architecture would replace the shared bus with a high-speed serial
link, with initial specification of 2.5 Gbps. The I/O subsystem would essentially
reside “outside” the server, within a message-passing, IP-based switched fabric.

258
The SAN Book III – October 22, 2001

The elegance of the InfiniBand model is twofold: 1) it minimizes CPU interaction


with storage processing, thereby reducing the burden on the server’s microprocessor,
and 2) it improves the scalability of the I/O subsystem.
The core technologies of InfiniBand are: 1) HCAs, 2) switching chips, and 3) TCAs.
Host Channel Adapters (HCAs) reside on the host, i.e. a server, sending data
through switches which create the interconnect to the Target Channel Adapters
(TCAs) which receive the data in target devices, such as storage subsystems and/or
InfiniBand-to-GbE routers. Given their relationship with server OEMs, HBA
vendors such as QLogic, Emulex, and JNI have entered the development arena for
InfiniBand technology. Silicon (i.e., chips) is being produced by established
semiconductor companies such as Intel and IBM as well as by startups such as
Mellanox, Banderacom, and others.
The development of industry standard software for InfiniBand (for subnet
management) is also being created by companies such as Lane15 and Vieo which
should help drive InfiniBand standardization and interoperability. The deployment
of industry standard subnet management software in a technology this young is
unique and has the potential to speed its adoption.
Figure 257 below illustrates where these products can be found on the network.
Note: Software is everywhere.
Figure 257. InfiniBand Network

Processor Node Processor Node

CPU CPU CPU

MEM HCA HCA MEM HCA

IB IB
Switch Switch
Storage Subsystems

TCA
IB
Network Router
TCA
Interface

Other IB
Network
Ethernet TCA subnets,
Interface
WANs

Fibre Channel

Source: Lane15

259
The SAN Book III – October 22, 2001

Figures 258 below illustrates current PCI and PCI-X architectures.


Figure 258. Today’s PCI and PCI-X Server Architectures

CPU CPU

Proprietary Internal Data Bus North Bridge &


South Bridge

Memory Controller PCI


Card

PCI Bus (66 MHz)

Source: Salomon Smith Barney

Figures 259 and 260 below illustrate potential InfiniBand architectures.


Figure 259. Phase 1: Potential InfiniBand Server Architecture

High-end

CPU CPU

Proprietary Internal Data Bus PCI Bus/


Proprietary Bus

Memory Controller

IB IB IB
HCA HCA HCA Multiple
Infiniband
Ports

Source: Salomon Smith Barney

260
The SAN Book III – October 22, 2001

Figure 260. Phase 2: Potential InfiniBand Server Architecture

Low-end Mid-range

CPU CPU
CPU
Proprietary Internal Data Bus PCI Bus/
Proprietary Bus
Memory Controller

IB Memory Controller
HCA

Multiple IB IB IB
Infiniband HCA HCA HCA Multiple
Ports Infiniband
Ports

Source: Salomon Smith Barney

InfiniBand Versus Existing Architectures


The InfiniBand ➤ Serial, Point-to-Point Connections: InfiniBand incorporates point-to-point
architecture has several connections on a switched fabric, unlike the current PCI architecture, which is a
compelling features that
make it more scalable
shared bus. The primary benefit of the point-to-point connection is that higher
and serviceable than bandwidth can be achieved through improvements in the media used to send the
existing architectures. signals. Each InfiniBand link will have two channels (1-bit wide), with a
theoretical signal rate of 2.5Gbps. (See estimated performance specifications in
Figure 261 below.) In addition to achieving higher bandwidths, point-to-point
connections do not require any interrupt processing to establish communication.
In contrast, the bus architecture requires that resources are shared, and various
prioritization schemes and interrupt processing must be determined.
➤ Single Point Upgrades: Because each InfiniBand link is independent, it can be
upgraded independently. In a shared bus (PCI or PCI-X) environment, the bus
speed drops to that of the slowest device.
➤ Improved Fault Isolation: Another benefit of the point-to-point architecture is
the ability to isolate faults to a single connection. A malfunctioning link can be
serviced without interruption of other I/O devices. On a shared bus, a faulty
device can bring down the entire bus segment.
Figure 261. InfiniBand Links and Data Throughput Rates
Link Link Signal Rate Useable Bandwidth Effective Throughput
Width (uni-directional) (80% of signal rate) (bi-directional)
1x 2-channel 2.5 Gbps 2 Gbps (250 MBps) (250 + 250) MBps
4x 8-channel 10 Gbps 8 Gbps (1 Gbps) (1 + 1) GBps
12x 24-channel 30 Gbps 24 Gbps (3GBps) (3 + 3) GBps
Source: Compaq Computer Corp.

261
The SAN Book III – October 22, 2001

Virtual Interface Enhances InfiniBand


Virtual Interface is a form Virtual Interface (VI) is often discussed in conjunction with InfiniBand architecture.
of Direct Memory Access It is a form of Direct Memory Access (DMA), a method to speed up the flow of data
to speed up the flow of
data from the application
from the application to the HBA. Benefits include low latency and reduced CPU
to the HBA. overhead due to its smaller instruction set. Emulex reports that database
applications have experienced greater than 40% improvement in performance when
utilizing its VI products.
VI is an architecture that sends data from node to node more quickly. It does this
through bypassing the Operating System (OS) layer and replacing other layers with
thinner ones that have less instructions. As you can see in Figure 262 below,
traditionally, once an application such as Excel generates data to send to the
network, the data must be processed by the OS (for example, Windows NT) and
then that information must be processed by the device driver (a software application
which manages a specific hardware piece) before it finally gets to the HBA and is
sent across the network.
Figure 262 below illustrates the VI architecture.
Figure 262. VI Architecture Comparison

Virtual
Standard Interface

Application Application
Operating System
Driver

HBA HBA

Source: Salomon Smith Barney

Using VI, the OS and the driver steps are bypassed, resulting in the data moving
more efficiently through the computer and out onto the network. In effect, VI
allows the data to take a shortcut from the application to the HBA. This is also
referred to as Direct Memory Access (DMA) since the data takes a direct path from
the main memory (application) to device memory (an HBA in this case) versus
going an indirect path by passing through additional steps. Note: In practice, the
driver is not completely eliminated but replaced by a thinner one with less
instructions.

262
The SAN Book III – October 22, 2001

VI Can Use Different Transport Protocols


VI can use Ethernet, VI only describes how data should be processed from the application to the HBA.
Fibre Channel, After those steps, other protocols are used to further prepare the data to be sent. VI
InfiniBand, or any one of
various protocols to
does not specify which transport protocol must be used to send the data - it can use
send the data. Ethernet, Fibre Channel, InfiniBand, or any one of various protocols.
Below we use the OSI model to illustrate how VI can work with a number of
network protocols.
Figure 263. VI in the OSI Model

Application
Presentation VI
Session

Transport
Network Fibre Channel,
Ethernet,
Datalink InfiniBand, etc.
Physical
Source: Salomon Smith Barney

Breaking Apart the Server Seems Logical


Back in time, when one monitor was connected to one PC with one keyboard and
one printer, computing was simple. Then computing moved to multiple PCs
connected to one printer. Then to multiple PCs getting data from one big server and
connecting to multiple printers, and so on. The difficult part is that each server can
only service so many PCs or applications. In order to expand one must either
increase the amount of servers or increase the robustness of each server. However,
one cannot just add microprocessors to servers, per se.
By using InfiniBand to Once a server is bought today, it has a finite capacity. By using InfiniBand to create
create 1U high 1U high microprocessor farms, servers can be consolidated and more efficiently
microprocessor farms,
servers can be
utilize floor space. We believe it creates seamless capacity expansion. This latter
consolidated and more point is very attractive to Internet Service Providers (ISPs) and Application Service
efficiently utilize floor Providers (ASPs) where floor space is in demand.
space.
Further, with the trend towards “bladed servers,” or literally, servers that are a single
card to be plugged in a chassis, InfiniBand becomes very powerful. A complete,
multiprocessor server can be built that is the size of a single card, and contains two
to four InfiniBand HCAs. With enabling software providing for what’s known as a
“shared I/O” paradigm, all traffic can share these HCAs into the fabric, be it Fibre
Channel storage traffic, Ethernet LAN traffic, or VIA server-to-server clustering
traffic. In this way, the cost of powerful computing nodes is driven down
tremendously.

263
The SAN Book III – October 22, 2001

Using VI over InfiniBand, servers can be aggregated much the same way that
storage utilizes RAID (Redundant Array of Independent Disks). For the server, we
will refer to this as RAIM (Redundant Array of Independent Microprocessors) See
Figure 264 below.
Figure 264. Potentially RAID and RAIM Architectures Could Look Very Similar

N
L AN

WA

N
LAN

WA
RAID Controller SAN SAN RAIM Controller
Fibre Channel

JBOD Microprocessor Farm (1U)

Infiniband
JBOD Microprocessor Farm (1U)

JBOD Microprocessor Farm (1U)

JBOD Microprocessor Farm (1U)

Source: Salomon Smith Barney

By disaggregating the components of a server, they can be scaled more efficiently


and provide greater reliability through any-to-any connectivity, thereby,
circumventing component failures and providing better resource utilization.
Figure 265 shows our view of the macro networking categories, including
InfiniBand. Note the overlapping areas, which are intentional. We expect the
overlapping portions to expand over time in the way we have depicted using arrows;
thereby creating competition between technologies.

264
The SAN Book III – October 22, 2001

Figure 265. Networking Diagram

WAN

LAN
SAN

IB

Source: Salomon Smith Barney

In an InfiniBand World...
Still in the concept stage, As part of the press surrounding InfiniBand, there has been a lot of noise about
InfiniBand won’t be InfiniBand eliminating the need for Fibre Channel and SANs. While it is true that
ready for deployment for
many years.
someday InfiniBand may challenge Fibre Channel for SANs given that the first-
generation InfiniBand specification has only recently been completed, we believe
the technology is still in its infancy and will take time to mature as a SAN
technology. Just as Fibre Channel was incubated for more than five years,
InfiniBand will also require several iterations before it’s ready for General
Availability deployment.
The Fibre Channel HBA Let’s fast-forward to 2003, when InfiniBand products are probably just starting to
suppliers will likely also gain momentum. At this time, Fibre Channel–based SANs and 10 Gb Ethernet
offer InfiniBand HCAs.
LANs will already be deep into deployment and often working in cohabitation.
InfiniBand will likely be deployed at the core of a data center, and connect to
existing LANs, WANs, and SANs. The question is then, which vendors will bring
InfiniBand products to market? Indeed, members of the server OEM, LAN, WAN,
and SAN communities have all been involved. We believe that given their close
working relationship with server vendors, our Host Bus Adapter suppliers, the same
ones offering Fibre Channel HBAs today, will be in the forefront of offering
InfiniBand host and target devices.
In short, we believe that everyone wants to use, and will use, InfiniBand.

265
The SAN Book III – October 22, 2001

Recent Company Developments

iSCSI HBA Product Roadmaps Announced


HBA vendors are beginning to offer iSCSI products. These devices propose to solve
a major problem with IP storage — the TCP/IP protocol stack is very CPU intensive.
New products will be able to offload the processing from the main CPU onto silicon
on the HBA.
Adaptec
On July 2, 2001, Adaptec acquired Platys Communications for $150 million. Platys
develops iSCSI ASICS that are scalable to 10 Gbps. On April 9, 2001, Adaptec
announced its AEA-7110C iSCSI HBA.
Emulex
On March 19, 2001, Emulex announced its GN9000/SI iSCSI HBA. It also offers
the GN9000/VI VI-over-IP host bus adapter for Virtual Interface.
QLogic
On January 23, 2001, QLogic acquired Little Mountain Group, a company that
develops iSCSI technology, for $30 million.
On July 31, 2001, QLogic licensed ARC International’s ARCtangent-A4 embedded
processor for its future iSCSI products.

Cisco Is Executing a Multi-Pronged Strategy


Cisco is attacking the Cisco is attacking the storage market on several different fronts. This multi-pronged
storage market on strategy involves working with 1) IBM, 2) Brocade, and 3) NuSpeed.
several different fronts
by working with 1) IBM, We believe Cisco is increasing its focus on storage networking because of the
2) Brocade, and 3)
market’s phenomenal growth potential. In Cisco’s words, “Storage Area Networks
NuSpeed.
(SANs) have become vital for companies to store and quickly retrieve data on
demand, as well as backup their systems.”
Further, there is speculation Cisco is incubating an in-house start-up called Andiamo
Systems, which is said to be developing a large-scale SAN Director that will
predominantly use iSCSI technology.

IBM
On February 21, 2001, IBM announced the IP Storage 200i, a NAS appliance that
uses the iSCSI protocol. IBM uses the same version of iSCSI (as Cisco) for its
product. Cisco and IBM jointly submitted the use of SCSI over TCP/IP to the
Internet Engineering Task Force (IETF), referred to as iSCSI.

Brocade
We believe the Cisco/Brocade partnership provides a very high level of validation
for the benefits and importance of Fibre Channel networking technology.
On June 13, 2000, Brocade announced a partnership with Cisco to interconnect
Cisco’s Catalyst 6000 product family of high performance multilayer switches with
Brocade’s Silkworm product family of Fibre Channel switches. The interconnection
will occur through the integration of a Fibre Channel–based interface card, code-

266
The SAN Book III – October 22, 2001

named Redhead, based upon Brocade’s e-port technology and located within a Cisco
switch, providing switch-to-switch connectivity for remote Fibre Channel SANs.
Cisco will pay Brocade a “license fee” on a per-port-shipped basis.
The combined solution is based on encapsulating the Fibre Channel protocol over
TCP stacked with IP through Ethernet to achieve non-blocking OC48 performance
(which is about 248 MBps, Fibre Channel currently runs at 200 MBps). Cisco will
continue to increase the speed of transmission with future generations of the
interface.
Cisco will also offer Cisco will also offer DWDM connectivity for Brocade to traverse Metropolitan Area
DWDM connectivity for Networks (MANs) and Wide Area Networks (WANs). (Note: ONI Systems is also
Brocade to traverse
Metropolitan Area
a Brocade partner and provides Brocade with DWDM. In addition, Brocade helped
Networks (MANs) and fund ONI before it went public.)
Wide Area Networks
(WANs). NuSpeed
On July 27, 2000, Cisco announced a definitive agreement to acquire NuSpeed, a
privately held company focused on IP based storage networking, for $450 million.
(The transaction closed in the quarter ended October, 2000.)
On April 9, 2001, Cisco announced the SN 5420 Storage Router, which utilizes
Cisco’s version of iSCSI. QLogic’s ISP2200A single-chip processor is being used
in the router. Some of the technology of the router came from the NuSpeed
acquisition.
Figure 266. IP SAN Router

Fibre
iSCSI Channel
Disk
Array

Router
Tape
Library
Source: Salomon Smith Barney

We believe NuSpeed has been focused on a two-pronged approach: 1) interconnect


Fibre Channel SANs using the Internet Protocol (IP), and then 2) use native IP to
network storage based on the iSCSI protocol. The first stage is expected to unfold
over the next few years; at which time, the second stage is expected to begin
ramping.

267
The SAN Book III – October 22, 2001

We believe this is an important announcement for iSCSI versus iFCP and FCIP. In
fact, Cisco’s announcement to acquire NuSpeed could have been an effort to stack
the deck in favor of Cisco’s iSCSI as the preferred IP storage networking standard.
This may seem to conflict with Cisco’s announcement that it was going to license
Brocade’s Fibre Channel technology to interconnect SANs; in fact, it does a bit in
one respect. With NuSpeed’s solutions (which, similar to Brocade’s technology, can
be embedded into a Cisco router and/or switch), Cisco will be able to interconnect
any Fibre Channel equipment through an IP network. With the Brocade technology,
Cisco will only be able to interconnect Fibre Channel equipment that conforms to
Brocade’s e-port standard (e-port is an inter-switch linking port). The
Cisco/NuSpeed router is more oriented to Fibre Channel node connectivity (i.e.,
storage, tape, etc.) versus Cisco’s other product, which connects to a Fibre Channel
SAN for long-distance connectivity.
Although we agree that efforts to develop IP based storage networks are gaining
momentum, we believe that IP-based storage networks will mainly service the lower
end and periphery of the market and will take at least two to three years to gain
traction in the enterprise. In the meantime, we believe that Fibre Channel will
continue to gain momentum as well.
Note: We do not expect Brocade to directly generate significant revenues from its
Cisco relationship. We do expect that Cisco will continue to develop and implement
Brocade’s e-port connectivity since it has been accepted as the Fibre Channel
interconnect standard by the American National Standards Institute (ANSI).
Separately, Brocade’s Fabric Shortest Path First (FSPF) routing protocol, a method
of routing data between various vendors’ Fibre Channel switches, was accepted as
the industry standard by the ANSI Technical Committee T11. (This is different, but
complementary, to Brocade’s e-port technology, which is the physical connection.)

What Has EMC Said and Done on the Subject?


EMC is agnostic, it just EMC is agnostic and just wants to sell more storage. We believe EMC will continue
wants to sell more to implement Fibre Channel SANs, while at the same time supporting an effort to
storage.
develop Ethernet as a server-to-storage interconnect. In fact, we believe that if using
spaghetti was the best way to network storage, EMC would fully support it. Note:
Both we and EMC agree that it will be several years before Ethernet can be reliably
used to network mission critical block-level storage traffic.

EMC Q&A:
Q: Will Ethernet be a viable server-to-storage interconnect?
A: “Yes, but not soon.” -EMC
Q: When will Ethernet technology be ready for prime time as a server-to-storage
interconnect?
A: “Two to five years. But Ethernet will not replace Fibre Channel. It will be
bridged into existing SANs.” -EMC

268
The SAN Book III – October 22, 2001

EMC’s Fibre Channel History


We believe EMC will We believe EMC helped ignite the “Fibre Channel versus Ethernet” debate back in
continue to implement December, 1995, when it bought McData (which priced its IPO on August 8, 2000,
Fibre Channel SANs,
while at the same time
at $28 per share and was subsequently spun out). McData was one of the first Fibre
support an effort to Channel storage networking companies (through its purchase of Canadian National
develop Ethernet as a Corporation from Hewlett-Packard). Through the first half of 2001 alone, EMC has
server-to-storage
generated approximately $1.75 billion of Fibre Channel–related revenues.
interconnect.
EMC’s Announcement of SRDF over IP Helped Ignite Today’s Debate
One of EMC’s most recognized developments has been its ability to remotely mirror
multiple Symmetrix storage subsystems in multiple locations in real time. The
benefit of this is the ability to recover from a disaster in which all systems in one
location fail; hence the term Disaster Recovery (DR). Lost data can cost companies
millions of dollars in current and future business.
Historically, remote mirroring has gone over leased T1 and T3 lines in order to
interconnect multiple sites (remember, IP is not currently robust enough to handle
block-level storage traffic effectively). EMC announced that its Symmetrix Remote
Data Facility (SRDF) is available over IP. So now IT managers can take advantage
of the Internet and cut costs by using SRDF over less expensive IP (Internet
Protocol) instead of having to lease expensive T1 and T3 lines right? Not quite.
(Note: IP is typically transported through Ethernet)
SRDF can go over IP, but not for Disaster Recovery because of the high latencies
associated with IP networking. EMC’s SRDF over IP only supports asynchronous
applications. In other words, you can only use SRDF over IP for static applications,
such as Internet content (an example of a non-static application would be
transactional data processing). We believe this actually opens up new opportunities
for EMC, which complements SRDF’s DR functions and, in turn, complements
EMC’s Fibre Channel investments.
EMC Has Entered the NAS Market to Offer Direct File Access over IP
A second application EMC is implementing using IP is NAS (i.e., attaching storage
directly onto Ethernet LANs). EMC’s NAS solution connects a Symmetrix or
Clariion storage subsystem directly to the LAN through its Celerra File Server.
EMC can also connect that same Symmetrix to a Fibre Channel–based Storage Area
Network (SAN) behind general purpose servers to keep block-level data transfers
separate, while at the same time utilizing the Symmetrix’s full capabilities. Notice
the convergence of storage networking behind general purpose servers (SANs) and
storage access directly from the LAN (NAS).

269
The SAN Book III – October 22, 2001

Figure 267. SAN and NAS Convergence with EMC’s Celerra


Tape
Library 1

Key Considerations
1)Relatively low cost of
Client 1
Server 1
management
EMC
2)Fault tolerant
Symmetrix 1 3)Data transmission speeds
Client 2
LAN EMC
Celerra 1 are somewhat constrained
by the LAN
EMC
Symmetrix 2 4)Heterogeneous
Client 3 N
SA connectivity
EMC
5)Increased scalability
Celerra 2 EMC
Client 4 Traditional Symmetrix 3
NAS
Appliance
Tape
Library 2

Source: Salomon Smith Barney

Figure 268. SAN, NAS, and DAS (Direct Attached Storage) Convergence with EMC’s Celerra

WAN/MAN/LAN
Tape
Library

Application
Server 1 EMC
Client 1 Sym 1

EMC
Client 2
LAN SAN Symmetrix 2
Application
Server 2
Client 3 EMC
Symmetrix 2
EMC Celerra
NAS Tape
Appliance NAS
Appliance 1

Source: Salomon Smith Barney

270
The SAN Book III – October 22, 2001

Figure 269. Total Storage Convergence Including Remote Mirroring


Tape
Library

EMC’s SRDF over


Dark Fiber
EMC
Client 1 Application Sym 1 EMC
Server 1 Symmetrix 4

EMC
Client 2
LAN SAN Symmetrix 2

EMC’s SRDF over IP


Application EMC
Client 3 Server 1 Symmetrix 5
EMC
Symmetrix 3 WAN/ MAN/
LAN/ SAN
EMC EMC’s EMC
Celerra SRDF Symmetrix 6
over IP

Source: Salomon Smith Barney

We believe EMC is anticipating a third application: Ethernet connectivity between


servers and storage (i.e., Ethernet-based SANs). At the same time, it is our
understanding that EMC sees storage networking traffic remaining separate from file
data traffic. In other words, even if Ethernet were just as robust as Fibre Channel,
storage and datacom networks would remain separate.
If the networks do, in If the networks do, in fact, remain separate, we believe this could benefit Fibre
fact, remain separate, we Channel storage networking, which has continued to gain momentum over that past
believe this could benefit
Fibre Channel storage
few years. We believe using Fibre Channel and Ethernet technologies to network
networking. storage will begin to overlap in the two- to five-year time frame.
EMC has told us that they do not foresee any precipitous change in the storage
networking landscape for the next couple of years. In other words, Fibre Channel
will continue to be the most robust block-level data storage networking technology
for the foreseeable future. Furthermore, the company does not have any preference
as to which technology customers will ultimately favor and we expect that EMC will
develop connectivity for both solutions, as long as it drives greater storage demand.
In the end, we believe EMC is investing in developing Ethernet connectivity (i.e.,
Ethernet SANs) as well as furthering its development of Fibre Channel connectivity
(i.e., Fibre Channel SANs).

271
The SAN Book III – October 22, 2001

This page intentionally left blank.

272
The SAN Book III – October 22, 2001

Coverage Universe

273
The SAN Book III – October 22, 2001

Companies in Our Universe

Advanced Digital Information Corp


Bell Microproducts
Brocade Communications Systems
Dot Hill Systems
EMC
Emulex
Hutchinson Technology
Inrange
Intevac
Iomega
JNI
Legato Systems
Maxtor
Network Appliance
OTG Software
QLogic
Quantum
Read-Rite
StorageTek
Veritas Software
Western Digital

274
The SAN Book III – October 22, 2001

7
Coverage Universe

Advanced Digital Information Corp. (ADIC, Storage


Subsystems)
Advanced Digital Information Corporation (ADIC) is a device-independent storage
solutions provider to the open systems marketplace. Through its recent Pathlight
acquisition, ADIC will be positioned to be a solutions provider for both tape and
disk, by our analysis. The company has been gaining market share and was the No.
2 and No. 3 OEM supplier of automated tape libraries in 2000 for units and
revenues, respectively. In 1999 and 2000, ADIC increased revenue share by 260
basis points and 270 basis points, respectively.
ADIC’s solutions span a wide range of organizational requirements with tape
libraries varying in size from seven to 70,000 cartridges. The company is drive-
independent and supports a broad range of drive types, including Quantum’s DLT,
Sony’s AIT, LTO Ultrium (IBM, Hewlett-Packard, and Seagate), and IBM’s
Magstar 3590.
All of the company’s products are designed to store, protect, manage and use
network data, including storage management software, Fibre Channel SAN
solutions, and NAS appliances.

Bell Microproducts (BELM, Distributor)


Bell Microproducts is a differentiated value-added provider of high technology
services, solutions, and products serving the industrial and commercial markets.
The company is a leader in the area of storage solutions and is the largest storage
distributor in the Americas.
Bell Microproducts differentiates itself by focusing on value-added services,
computer storage and related peripherals, and niche semiconductor products. Value-
added services include storage design, integration, implementation, and support;
subsystem testing; software loading; and disk drive formatting and testing.
In April, 2001, Bell Microproducts was added to the Fortune 1000 list and was
ranked No. 759 based on 2000 revenues.

Brocade Communications Systems (BRCD, Storage


Networking)
Brocade Communications Systems is the leading supplier of Fibre Channel fabric
switch solutions, which provide the intelligent backbone for SANs. The Brocade
SilkWorm family of Fibre Channel switches and Brocade SAN management
software are designed to optimize data availability in the storage enterprise. Using
Brocade solutions, companies can simplify the implementation of SANs, reduce the
total cost of ownership of data storage environments, and improve network and
application efficiency and performance.
Brocade has been a leader in the SAN networking industry since the company’s
inception in 1995. Brocade has consistently offered leading-edge products, which
have consistently satisfied customers’ evolving storage networking needs.

275
The SAN Book III – October 22, 2001

Dot Hill Systems Corp. (HIL, Storage Subsystems)


Dot Hill Systems Corporation is an independent provider of storage and SAN
solutions and services. The company designs, manufactures, markets, and supports
data storage systems for open systems computing environments, including Unix,
Windows, and Novell.
Dot Hill’s disk-based product line is called SANnet and is sold primarily into SAN
environments. The company currently has approximately 250 employees. Dot
Hill’s major offices are located in New York City; San Jose; Washington, DC;
London; Tokyo; Beijing; Germany; Holland; and Israel. Dot Hill’s primary
manufacturing facility is located in Carlsbad, California.
Dot Hill is the result of a merger between Box Hill and Artecom in August, 1999.
Box Hill was focused on providing storage solutions to New York’s financial
service institutions, while Artecom was angled toward the telecom sector.

EMC (EMC, Storage Subsystems)


EMC is the largest and most diverse pure play data storage infrastructure storage
company in the industry. The company has gained its market-leading position as a
result of aggressive sales efforts, a keen understanding of customers’ needs,
comprehensive service support and the delivery of solid, consistent, reliable
products that offer one of the greatest feature sets in the industry.
We believe EMC has strategically positioned itself as a leading company in every
storage segment it has entered, including: enterprise storage hardware, storage
software, networking, and the high-end Network-Attached Storage (NAS) market.

Emulex (EMLX, Storage Networking Components)


Emulex is a leading supplier of Fibre Channel host adapters, hubs, ASICs, and
software products that enhance access to, and storage of, electronic data and
applications. The company has exited its traditional networking business and de-
emphasized its hub business in an effort to focus on the opportunities within Host
Bus Adapters. Focused on high performance and a rich feature set, Emulex has
achieved much of its success in the premium segment of the fibre channel host bus
adapter (HBA) market.

Hutchinson Technology Inc. (HTCH, Disk Drive


Components)
Hutchinson Technology is the world’s leading manufacturer and supplier of
suspension assemblies for hard disk drives (market share of between 45% and 50%).
Hutchinson manufactures conventional suspension assemblies, TSA suspension
assemblies, and suspension assembly components. Its fiscal 2000 revenue mix was
22% conventional suspensions and 76% TSA suspensions.
Hutchinson’s products are sold to disk drive manufacturers (original equipment
manufacturers) for use in their products and to head manufacturers (sub-assemblers),
who sell to original equipment manufacturers.

276
The SAN Book III – October 22, 2001

In second quarter 2001, enterprise hard disk drives accounted for over one-third of
suspension production, desktop drives accounted for about 50%, and mobile hard
drives accounted for the remainder.
Hutchinson works closely with its customers to develop products that improve disk
drive performance and reduce manufacturing cost.

Inrange (INRG, Storage Networking)


Inrange is a storage networking company with a branded end-to-end total storage
networking solution. Inrange will install and certify all storage networking
equipment, including HBAs, 16-port switches, 128-port Directors, Dense Wave
Division Multiplexing, and Fibre Channel-to-Gigabit Ethernet and Fibre Channel-to-
ATM routing.
Inrange has entered the forefront of the storage networking industry by leveraging
its previous expertise in storage and datacom and telecom networking to develop
one of the industry’s most robust and technologically advanced solutions. Its
FC/9000 128-port Fibre Channel Director has the highest port count in the industry
today. Inrange is also uniquely positioned as the only storage networking company
with robust in-house, direct sales and services organizations. We believe that both
of these in-house organizations will be significant contributors to Inrange’s overall
storage networking success. Inrange is exiting its telecom businesses to increase its
focus as a storage networking company.

Intevac (IVAC, Disk Drive Capital Equipment)


Intevac (IVAC) is a leading supplier of sputtering systems and related equipment
used to manufacture thin-film disks for computer hard disk drives. Intevac has also
been pursuing other businesses (including flat panels).

Iomega Corporation (IOM, Disk Drives)


Iomega Corporation designs, manufactures, and markets personal and professional
storage solutions that are based on removable-media technology for users of
personal computers and consumer electronics devices. The company’s primary data
storage solutions include disk drives and disks marketed under the trademarks Zip,
Iomega CD-RW, Jaz, PocketZip, and Peerless high capacity storage. Iomega is the
dominant supplier in the low-end removable magnetic market with its Zip products
and participates in the high-end magnetic markets with its Jaz and Peerless product
lines.

JNI (JNIC, Storage Networking Components)


JNI is a leading supplier of Fibre Channel hardware and software products that
connect servers and data storage devices to form SANs. The company currently
provides high-performance ASICs, a broad range of Fibre Channel HBAs, and
software that facilitate the integration and management of the devices that make up a
SAN.
The company has a solid foothold in the non-captive Solaris (Solaris is Sun
Microsystems’ platform) market, and benefits from relationships with major storage

277
The SAN Book III – October 22, 2001

OEMs such as EMC, Compaq, and StorageTek. Most of JNI’s success to date has
been from connecting Sun servers to storage subsystems from EMC, Compaq,
StorageTek, Hitachi Data Systems, and others. Given the stability and popularity of
Sun’s Solaris platform, JNI has benefited tremendously from its growth. The
company is now building on its Solaris success, and expanding into other UNIX-
based architectures such as HP-UX and IBM’s AIX.

Legato Systems (LGTO, Storage Software)


Legato Systems develops products in the enterprise storage management software
market. Legato helps customers leverage business-critical and corporate data assets
through its storage management software. Legato’s products enable information
continuance, and the management, movement, and protection of data throughout an
enterprise. Legato’s products are available through a network of Legato-licensed
value-added resellers, resellers, and OEM partners.
Legato’s strategy is to enhance hardware vendors’ product offerings by providing
advanced software applications, thereby obtaining and leveraging hardware vendors’
support and market power to help Legato further penetrate the storage management
software market. Legato achieves this goal by 1) staying clear of the file system
product category, 2) writing to a specifically defined set of Application Program
Interfaces (APIs) which are publicly available, and 3) developing software
applications that offer integrated and heterogeneous solutions.

Maxtor (MXO, Disk Drives)


After the merger with Quantum-HDD in April, 2001, Maxtor is now the largest hard
disk drive (HDD) maker and manufactures one of the broadest product lines in the
disk drive industry. In addition to HDDs, Maxtor manufactures workgroup/entry
NAS devices under its MaxAttach label.
In second quarter 2001, Maxtor’s desktop drives accounted for 89.4% of revenues
and 96.4% of units, while its enterprise segment accounted for 9.7% of revenues and
3.6% of units. Maxtor sold 60% of its products through OEMs, 37% through
distributors, and 3% through retail. Geographically, Maxtor sold 45% domestically,
25% in Europe, 27% in Asia, and 3% in Latin America and Canada. In second
quarter 2001, NAS revenues accounted for 0.9% of total revenues.

Network Appliance (NTAP, Storage Subsystems)


In 1992, Network Appliance helped originate Network Attached Storage (NAS).
Although Network Appliance may not have been the sole inventor of the NAS
concept, it has been one of the marquee NAS companies. Network Appliance also
offers a family of Web-caching products, but a large majority of its revenues are
currently derived from its family of file servers for NAS.
Network Appliance’s strategy is to remove the Input/Output (I/O) burden of file
accessing from application servers with a separate dedicated file storage appliance
that connects directly onto the Local Area Network (LAN). Network Appliance’s
NAS products, which it calls Filers, are essentially a high-speed front end to large
numbers of disk drives. Today, these Filers can hold as much as six terabytes of

278
The SAN Book III – October 22, 2001

information (nine terabytes in the near future). NetApp Filers feature built-in RAID,
clustered failover and redundant components for increased reliability.

OTG Software (OTGS, Storage Software)


OTG Software is an emerging storage software player focused on storage
management and access which leverages its industry leading Hierarchical Storage
Management (HSM) technology. HSM is intelligent software that migrates unused
or less-used data from primary storage subsystems to less expensive secondary
storage subsystems, enabling the expansion of accessible storage capacity without
the burden or expense of installing a new primary storage subsystem. OTG has
leveraged its core HSM technology by adding many data management and access
tools, such as content management and search/archive functionality.
We see the company’s EmailXtender as one of the leading “killer applications” in
storage and believe it will increase OTG Software’s visibility with customers and
leverage it into broader customer relationships, including the sale of other Xtender
solutions.

QLogic (QLGC, Storage Networking Components)


QLogic is the only supplier of end-to-end Fibre Channel solutions which include
HBAs, integrated circuits (ICs), and switches. We believe QLogic’s chip business is
a clear leader in the industry and see its Ancor switch acquisition beginning to pay
off nicely.
While Emulex targets high-end UNIX systems and JNI targets S-bus systems for
after-market attach to Sun servers, QLogic has become the low-cost supplier of
Fibre Channel solutions by building inexpensive, highly integrated products. Top
customers include Fujitsu, Sun, Dell, and Network Appliance.

Quantum (DSS, Storage Subsystems)


Quantum designs, develops, and markets information storage products, including
half-inch cartridge tape drives (DLT), tape media, tape autoloaders and libraries, and
entry/midrange NAS subsystems.
Quantum’s fiscal 2001 (March) revenue mix was 47% tape drives, 24% media (tape
cartridges), and 29% storage systems (tape libraries, NAS, storage solutions).
Quantum’s DLTtape drives are the dominant tape drive in the mid-range tape market
as Quantum recently introduced its next generation tape drive called SuperDLT. In
October 2000, Maxtor acquired Quantum’s hard disk drive operations (Quantum-
HDD).

Read-Rite Corporation (RDRT, Disk Drive Components)


Read-Rite Corporation is one of the world’s leading independent manufacturers of
magnetic recording heads, head gimbal assemblies (HGAs), and head stack
assemblies (HSAs) for disk drives and tape drives. The company is the only U.S.-
based independent supplier of magnetic recording heads for the computer storage
industry. In addition to its core read-head business, Read-Rite currently has

279
The SAN Book III – October 22, 2001

approximately 73% ownership in Scion Photonics, a supplier of optical filters and


waveguide, and contract manufacturer.
Read-Rite is headquartered in Fremont, California, and has operations in Japan, the
Philippines, Singapore, and Thailand. Major customers include Maxtor, Samsung
and Western Digital. Read-Rite’s two competitors in the OEM read-head market are
Alps Electric and TDK.

StorageTek (STK, Storage Subsystems)


StorageTek designs, manufactures, markets, and services information storage and
retrieval subsystems and networking products. StorageTek is focused on delivering
information storage solutions for the mainframe and client/server market, including
UNIX and WindowsNT environments. The company’s strategic business segments
are tape and tape automation, disk, and networking. For 2000, tape and tape
automation accounted for 54% of revenues; disk, 7%; networking, 8%; and service,
31%.

Veritas Software (VRTS, Storage Software)


Veritas Software is a leading enterprise-class application storage management
software provider that ensures information availability from business-critical
applications by delivering integrated, cross-platform storage management software
solutions. Veritas positions its products as 1) Backup, 2) File and Volume
Management, and 3) Clustering/Replication. The company’s products enable
business to continue without interruption and are designed to protect, access and
manage business-critical application information.
Veritas has been able to outpace overall market growth both by acquiring niche
companies, thereby entering new market segments, and offering a total software
solution. A significant portion of Veritas’s above-average growth is the result of the
current trend by customers to purchase storage management software solutions from
a single vendor that can provide a more complete solution. There are few other
companies that can offer a “total solution.”

Western Digital (WDC, Disk Drives)


Western Digital supplies hard disk drives to desktop and consumer applications. For
calendar 2000, Western Digital’s revenues were $2.0 billion, with desktop drives
accounting for virtually 100% of revenues by the fourth quarter.
In early 2000, the company announced that it was exiting the enterprise segment and
has been investing heavily in its subsidiaries: Cameo (communication and network
applications), Keen PM (consumer electronics), SageTree (supply chain
management), SANavigator (SAN software). In calendar 2000, Western Digital
held 10.7% of total HDD unit market share with 14.4% of desktop market. Western
Digital sold 70% of its products through OEM, 22% through distributors, and 8%
through retail. Geographically, Maxtor sold 57% domestically, 30% in Europe, 13%
in Asia.

280
The SAN Book III – October 22, 2001

Private Companies

281
The SAN Book III – October 22, 2001

Listing of Private Companies

3PARdata NSI Software


3ware nStor
Alacritech OmegaBand
Ancot Panasas
Aprilis Pirus
Astrum Software PolyServe
ATTO Technology PowerQuest
BakBone Software Prisa Networks
Banderacom Raidtec
Baydel RedSwitch
BlueArc Rhapsody Networks
Broadband Storage SanCastle Technologies
Bus-Tech Sanera Systems
Cereva Networks SANgate Systems
Chaparral Network Sanrise Group
CMD Technology SAN Valley Systems
CommVault Systems Scale Eight
CreekPath Systems Seagate/XIOtech
Datacore Software Shugart Technology
DataDirect Networks Storability
Eurologic Systems StorageApps
ExaNet StoreAge Networking
Hitachi Data Systems Tantia Technologies
I-TECH Times N Systems
Ikadega Troika Networks
InfiniCon Systems TrueSAN Networks
Infiniswitch Viathan
InterSAN Vicom Systems
Lane15 Software VIEO
LeftHand Networks Voltaire
ManagedStorage YottaYotta
Mellanox Technologies Zambeel
NetConvergence Zantaz
Nishan Systems

282
The SAN Book III – October 22, 2001

7
Private Companies
Note: Companies are listed in alphabetical order. The length of individual
company profiles is based on the availability of information and not on preference.
For example, companies we have previously published on typically have longer
profiles, while companies in stealth mode have shorter ones.

3PARdata Inc. (Storage Subsystems)


3PARdata, founded in May, 1999, builds Utility Storage Servers and software that
enable information intensive organizations to consolidate and centralize their
information and storage assets so that they can be exploited to their full potential.
3PARdata Utility Storage Servers enable multi-tenancy, management automation,
and new levels of performance and scalability.
3PARdata, still in stealth mode, develops Carrier-Class Storage Servers that manage
and serve information so that information consumers can fully exploit it. There is a
great need for simplified storage solutions that scale in performance, capacity, and
connectivity, and provide intelligent and flexible management software for
demanding big data applications.
Carrier-Class Storage Servers are a new category of storage solutions designed for
Merchant Service Providers, known today as xSPs and Enterprise Service Providers,
that provide:
➤ non-stop information services with failure tolerance, non-disruptive hardware
and software upgrades, minimal performance degradation under failure, and
proactive and predictive support;
➤ automated management with virtual capacity allocation, usage tracking, granular
quality of service, performance optimization, and simple volume and raid
management that eliminate server-based storage management; and
➤ an information consolidation engine offering true scalability in performance,
capacity, and connectivity.
The company has strategic investors that include Oracle, Sun Microsystems, and
Veritas. Financial investors include: Amerindo, Anshutz, Aurora Funds, Dain
Rauscher Wessels, IGlobe Partners, Intec IT Capital, Mayfield, and Worldview
Technology Partners. 3PARdata most recently raised $100 million in June, 2001.

3ware (Storage Subsystems and Components)


Founded in February, 1997, 3ware has developed a new storage I/O architecture that
raises the performance and lowers the cost of storage by utilizing existing
infrastructures and technologies such as Ethernet and IDE/ATA drives. We believe
3ware is well positioned to penetrate the expanding lower-cost storage market.
3ware utilizes IDE/Ultra ATA drives in order to reduce cost and complexity, but
without compromising performance or reliability. The company’s technologies
include DiskSwitch, AccelerATA, and TwinStor.
At the core of 3ware’s architecture is its Network Storage Unit (NSU) Storage
Subsystem, which, when combined with its Escalade family of Switched Storage

283
The SAN Book III – October 22, 2001

RAID controllers and 3DM Remote Storage Management software, creates one of
the most promising emerging low-cost storage solutions. The 3ware solution allows
the transmission of block-level SCSI data over Gigabit Ethernet LANs or subnet
SANs, enabling customers to leverage existing investments in network
infrastructures, reduce management complexity, significantly lower overall costs,
and potentially eliminate the need for Fibre Channel IT personnel. The 3ware NSU
is available in configurations from 240GB to 600GB and supports Windows 2000,
Windows NT, MacOS, Solaris, and Linux platforms.
3ware’s Escalade controllers, which are packaged with the NSU, are based on the
company’s DiskSwitch packet-switching technology. 3ware developed its
DiskSwitch Architecture to decrease latency and increase performance. The
DiskSwitch storage controller architecture applies the principles of packet-switched
networking to subsystem I/O. Each disk drive has fully dedicated bandwidth,
eliminating bus arbitration delays and other limitations of shared-bus architectures.
Low-cost UltraATA drives are used by the company in place of SCSI or Fibre
Channel drives, and the company claims this is without incurring any performance
penalty. Escalade controllers are available in two-, four-, and eight-port models and
can be sized to fit any configuration from a 1U rack-mounted system to a high-
capacity, multi-controller system.
3ware’s 3DM Remote Storage Management software displays drive and RAID array
status, allows remote configuration of new arrays, and enables hot swapping of
failed drives for on-the-fly rebuilding. 3DM’s Web-based user interface provides a
common look and feel for all platforms, and, because it is browser-based, customers
can manage their storage from any point of network access without installing
software on the remote system.

Alacritech (Storage Components)


Founded in 1997, Alacritech produces fully scalable network server adapters that
offload network protocol data processing from the central processing unit on NT
servers. Alacritech manufactures SLIC (Session-Layer Interface Card) Technology
to accomplish this task by using an application-specific integrated circuit (ASIC)
and Alacritech’s custom Internet Protocol Processor (IPP). SLIC is fully compatible
with Ethernet, TCP/IP, contemporary PC hardware and software, as well as today’s
legacy network adapters and existing networking infrastructures.
Alacritech’s SLIC technology is designed to enable enhanced Gigabit Ethernet
performance by offloading TCP processing from networking endpoints, allowing
systems to drive many gigabits per second of information at greater efficiencies with
boosted data throughput and decreased latency. Alacritech manufactures single-,
two- and four-port 10/100 server adapters that conform to PCI 2.2 specification for
64-bit devices for operation in either 32-bit or 64-bit PCI environments. These
server adapters allow IT departments to schedule more backups in the same amount
of time. Alacritech is currently testing its Gigabit Ethernet Server Accelerator, with
availability planned for the end of 2001.
Alacritech products currently support iSCSI. Cisco Systems and FalconStor have
demonstrated iSCSI operating on standard Alacritech Server Accelerators, and

284
The SAN Book III – October 22, 2001

Alacritech intends to offer iSCSI adapters in late 2001. Additionally, the company
has started developing a 10 Gb Ethernet solution. Depending upon completion of
the standard, it plans to offer this product in 2003.
Figure 270. 100x4 Quad-Port Server Accelerator

Source: Company website

Figure 271. 100x1 Single-Port Server Accelerator

Source: Company website

In June, 2001, Alacritech announced its 100 Series Server Accelerators support MS
Windows 2000 and Redhat Linux operating systems. Alacritech’s founder, Mr.
Larry Boucher (inventor of the SCSI interface) has previously founded two
successful technology companies, Adaptec (ADPT) and Auspex Systems (ASPX).
The company is privately held and has completed two rounds of funding totaling
$22.4 million from Benchmark Capital, Institutional Venture Partners, Alloy
Ventures, and Berkeley International.

Ancot Corp. (Storage Testing)


Ancot designs and manufactures a family of UltraSCSI and Fibre Channel test
instruments and associated equipment, such as bus analyzers and converters,
targeted at developers and system integrators. Its customers are companies that
develop, offer, and use the SCSI and Fibre Channel interface for their computer,
peripheral, and network end-user products. Its markets include all major computer
companies, peripheral manufacturers of disk, tape, CD-ROM, WORM, systems

285
The SAN Book III – October 22, 2001

integrators, value-added resellers, and SCSI computer end users, as well as network
equipment suppliers.
Ancot was founded in 1987 by President Jan Dedek, to provide SCSI test and
development equipment to the microelectronics industry. Since then, Ancot’s
products have evolved, keeping pace with the growth and changes of the SCSI bus
protocol. Ancot shipped the industry’s first Fibre Channel analyzer in 1994 and
continues to develop SCSI and Fibre Channel products. Combining SCSI and Fibre
Channel technologies, Ancot now offers a more complete line of analyzer and
extender solutions.
For its upper tier customers, Ancot offered the industry’s first 2 Gb protocol
analyzer for Fibre Channel systems. Ancot also offers a server version Fibre
Channel Analyzer that supports 1.06 Gbps data rates, yet cuts in half the typical cost
for advanced Fibre Channel instrumentation.
➤ FCAccess 2000 Fibre Channel Analyzer: supports both 1.06 Gbps and 2.12
Gbps data transfer speeds and is a leading network-attached, client-server Fibre
Channel Analyzer.
Figure 272. Ancot — FCAccess 2000 Fibre Channel Analyzer

Source: Company website

➤ The Ultra2160 and Ultra2160/Lite SCSI Bus Analyzers: premier tools for
debugging and optimizing SCSI devices and systems.
Figure 273. Ancot — The Ultra2160 SCSI Bus Analyzer

Source: Company website

The company’s world headquarters is located in Menlo Park, California. Sales and
support for U.S. customers are provided from this location, while international
markets are reached through its worldwide distributor network.

286
The SAN Book III – October 22, 2001

Aprilis Inc. (Storage Subsystems)


Aprilis, located in Cambridge, Massachusetts, commenced operations in June, 1999.
It was established to commercialize holographic data storage technology exclusively
licensed from Polaroid Corporation, with an early focus on large emerging markets
in three-dimensional data storage for ultra high capacity and high performance
applications. Government funding through Defense Advanced Research Projects
Agency–sponsored consortia supported early development of a proprietary
holographic recording media based upon cationic ring opening polymerization
(CROP).
Aprilis is currently participating in activities in support of the PRISM consortium, a
five-year DARPA-funded effort that has investigated a number of media options for
use in a prototype holographic storage system designed by HDSS, PRISM’s
companion consortium. Polaroid’s membership in PRISM began in 1997 and its
high performance photopolymer was rapidly selected to be the write-once-read-
many (WORM) recording material for the HDSS system demonstration. The HDSS
system specifications comprise a capacity of one terabit on one disk, random access,
and a sustained data rate of six gigabits per second using a single optical head.
Additionally, Aprilis is currently collaborating with several leading independent
developers of holographic data storage technology, both in the United States and
abroad, by meeting their stringent requirements for high-performance WORM
recording media.

Astrum Software (Storage Software)


Astrum Software, a privately held company based in Boston, Massachusetts,
delivers client-to-storage management solutions for Windows 2000/NT/9X based
systems, storage area networks and network attached storage. Astrum’s product
suite, StorCast, enables customers to proactively manage explosive storage growth,
ensuring accessibility and availability to an organization’s most critical assets —
data and IT infrastructure.
Astrum’s StorCast Suite is a family of fully integrated storage resource management
software products that provide a centralized view to manage the implementation and
growth of storage in today’s distributed RAID, SAN, and NAS environments. The
StorCast Suite includes the Storage Asset Manager, Reporter, Interconnect Monitor,
and Storage Performance Manager.
➤ StorCast Storage Asset Manager: The quickest and easiest way to control
storage consumption, monitor system performance and perform trend analysis.
Supports Windows 2000/NT/9x.
➤ StorCast Reporter: A quick and easy way to discover all storage resources
throughout your IT infrastructure. Automated storage reporting and forecasting
solution. Supports Windows 2000/NT/9x.
➤ StorCast Interconnect Monitor: A powerful, simple, and scalable solution for
verifying connectivity and availability of systems, devices, ports, and network
attached storage (NAS) throughout the enterprise and the Internet from a central

287
The SAN Book III – October 22, 2001

location. Supports Windows 2000/NT/9x, Linux, AIX, NetWare, OpenVMS,


HP-UX, UNIX.
➤ StorCast Storage Performance Manager: Provides a storage-centric view of
enterprise performance, ensuring all system and network components operate at
peak productivity.

ATTO Technology, Inc. (Storage Networking)


Headquartered in Amherst, New York, ATTO Technology designs and
manufactures host adapters, SAN bridges, hubs, routers, and offers complete SAN
solutions. Established in 1988, ATTO Technology develops a broad range of SCSI
and Fibre Channel products that enable users to accelerate data access, increase
business productivity, and better manage the storage of enterprise information
through SAN development.
Since its inception, ATTO Technology has focused on a single goal: faster and more
flexible data delivery and productivity to maximize total cost of ownership. From
host adapters, to bus expansion and connectivity devices, to solid state storage,
ATTO Technology’s peripheral products are designed to improve response time and
optimize system, and in particular, storage network performance, for computer users
around the world.
ATTO Technology’s hardware and software solutions suit the needs of high-end
workstations such, as Sun Microsystems, DEC, Hewlett-Packard, and Apple
platforms. The company designs products for specialized applications including
digital content creation, pre-press/electronic publishing, Internet/ISPs, data mining,
decision support, asset management, online transaction processing, imaging,
CAD/CAE, and software development. ATTO Technology implements fully tested,
custom SAN solutions through its ContentCreation Solutions SAN package and
serverless backup utilizing Fibre Channel Technology. To test its products and
complete storage solutions, ATTO Technology has built a state-of-the-art
interoperability lab that enables testing solutions for interoperability between SCSI,
InfiniBand, and Fibre Channel.
ATTO Technology was the first to introduce and/or to ship a:
➤ SCSI-based stand-alone hardware cache (ATTO SiliconCache);
➤ SCSI Logical Unit Translator product (ATTO SCSI Expander), which expands
the number of SCSI devices that can be put on a single SCSI bus;
➤ BusMaster SCSI card for Macintosh;
➤ Dual-channel (ATTO ExpressPCI UL2D) Ultra2 host adapter, as well as single
and multichannel PCI-based SCSI-3 host adapters, all of which can be used on
either Macintosh or PC systems without the need for reconfiguration.
In May, 2001, ATTO announced that its ExpressPCI FCSW HBAs have been
qualified for use on “Powered by DataCore” storage virtualization nodes. As part of
DataCore’s SANvantage Technical Partner Program, ATTO and DataCore ensure
the use of these Fibre Channel host bus adapters as target and initiator ports for
SANsymphony-managed network storage pools. The ExpressPCI FSCW enables

288
The SAN Book III – October 22, 2001

PCI-to-Fibre Channel interconnect while maximizing productivity in workgroup and


enterprise-level SAN environments. It utilizes an embedded Short Wave Laser
Interface, capable of streaming data over cable lengths of up to 500 meters. The
adapter features full duplex 1.0625 Gb data-transfer rates, Class 2 and F-Port
support, 64/32-bit PCI bus interface and 66/33 MHz compatibility. By integrating
ATTO ExpressPCI HBAs into SANsymphony storage virtualization nodes,
substantial improvements in I/O rates translate into greater accessibility and
availability of data, according to the company.
ATTO Technology’s Enterprise solutions are ideal for high demand computing
environments where data reliability, availability, and serviceability are critical.
Enterprise solutions provide users with the tools necessary to maximize network
uptime. ATTO Technology provides the building blocks necessary to take
advantage of Enterprise SANs, implement LAN-free backup and take advantage of
Serverless backup.
Figure 274. ATTO Technology’s Enterprise Solutions

Source: Company website

The ATTO ExpressPCI FC 3305 brings 2 Gb throughput to server cluster and SAN
environments. The FC 3305, with a fixed copper HSSDC Fibre Channel interface,
allows for high-speed copper connections. Designed to support next-generation
SAN requirements, the FC 3305 is ideal for high-speed random or sequential types
of data-intensive environments.

289
The SAN Book III – October 22, 2001

Figure 275. ATTO ExpressPCI FC 3305

Source: Company website

ATTO FibreBridge 3200R is an intelligent, managed Fibre Channel-to-SCSI bridge


with Advanced Fibre Channel Features designed for 24x7, high reliability, and high
availability Enterprise applications. Interoperability capabilities are amplified with
the addition of support for F-Port and full duplex transfers, along with dual GBIC
ports and dual SCSI ports available in an LVD option. By achieving 98 MBps of
sustained throughput, the FibreBridge 3200R expands the speed potential of
Enterprise applications.
Figure 276. ATTO FibreBridge 3200R

Source: Company website

The ATTO FibreCenter, Fibre Channel hub serves as a high-speed, central


connection point for data sharing in Fibre Channel Arbitrated Loop (FC-AL)
configurations. This stackable, five-port hub provides full duplex 1.0625-Gigabit
performance with 200 MBps transfer rates for I/O and data intensive Content
Creation environments.
Figure 277. ATTO FibreCenter, Fibre Channel hub

The ATTO FibreCenter 2100R serves as a high-speed, central connection point for
data sharing in Enterprise Fibre Channel Arbitrated Loop (FC-AL) configurations.
The addition of a rack-mounted Fibre Channel hub in the FC Rack System gives
Enterprise users countless options for configuring, upgrading, and expanding Fibre
Channel Storage Area Networks (SANs). This five-port hub provides full duplex

290
The SAN Book III – October 22, 2001

1.0625 Gb performance and delivers 200 MBps transfer rates for I/O and data-
intensive SAN environments.
Figure 278. ATTO FibreCenter 2100R

Source: Company website

Worldwide distribution of ATTO products is directed through OEMs, SIs, VARs,


and major distributors such as Bell Microproducts, Consan, and Ingram Micro.
Customers include Apple, EMC–Data General, Eastman Kodak, Quantum/ATL, Sun
Microsystems, Avid Technology, and others.

BakBone Software (Storage Software, Not Rated)


BakBone Software (BKB.To) is a storage management software solutions company
with offices in the Unites States, the UK, Japan, and Canada. The company
develops and distributes storage management software solutions to the open systems
markets. It provides data protection and management solutions scalable from office
to enterprise that will grow with a company’s needs.
Building upon core technology originally developed at AT&T/Bell Labs, BakBone’s
NetVault software products are designed to provide a robust data protection and
recovery solution. NetVault’s features include versatile modular design, SAN/LAN-
free support and specialized online application backup. Currently, NetVault
supports both Windows NT and UNIX environments with support for Windows
2000 planned for the future. Customers include major server OEMs and storage-
centric value added resellers and distributors worldwide.
In June, 2000, BakBone acquired Tracer Technologies, now considered a key asset
in BakBone’s future. Tracer’s MagnaVault is a market leader in file systems and
device drivers for optical storage systems, with customers ranging from global
financial institutions to major agencies of the United States Government.
BakBone has a strategic alliance with FalconStor Software, a network storage
infrastructure company that provides worldwide bundling of BakBone’s NetVault
backup/restore software with FalconStor’s IPStor. Customers are able to run
NetVault on IPStor servers to locally backup and restore their data, while at the
same time relieving application servers on an IP-based SAN from the processing
burdens imposed by traditional backup methods. Current users include AT&T,
Bank of America, Federal Express, Cable & Wireless, Daimler-Chrysler, NCR, SUN
Microsystems, Sony, and others.
BakBone is located in San Diego, California and its facility provides office space, an
interoperability certification and test lab, and a global training and education facility
for employees and distribution partners.

291
The SAN Book III – October 22, 2001

Banderacom (InfiniBand Networking)


Banderacom is a privately held, fabless semiconductor company located in Austin,
Texas, focused on providing I/O connectivity silicon for computing and embedded
systems environments utilizing the InfiniBand standard. Banderacom develops
highly integrated semiconductors for InfiniBand host, targets, and switching
products.
Banderacom’s IBandit extensible architecture provides the building blocks for
constructing the InfiniBand fabric infrastructure of the data center. IBandit allows
rapid development of InfiniBand connectivity solutions to bridge the gap between
legacy PCI I/O and InfiniBand. Whether applications demand Fibre Channel, SCSI,
SONET, Gigabit Ethernet, or other protocols used in applications such as I/O
chassis, RAIDs, and JBODs, Banderacom’s system-on-a-chip takes advantage of
InfiniBand’s high bandwidth and management features.
Founded in November, 1999, as INH Semiconductor, Banderacom’s founders were
all senior members of the technical staff at Jato Technologies, a networking silicon
startup now owned by Intel Corporation.

Baydel (Storage Subsystems)


Baydel believes Ethernet SANs, not Fibre Channel SANs, are the wave of the future.
This contrast with many other SAN solution providers is gaining momentum with a
number of startups and industry giants as well. Baydel designs and integrates
computer storage, connectivity, and system products. Very proud of its in-house
research and development teams, Baydel’s design capabilities encompass circuit
design, microcode and software, and mechanical and electromechanical assemblies.
The O/ESR 2000 is a significant new RAID storage system from Baydel which
incorporates all the well proven features of the current O/ESR(Raider-5) product and
adds an impressive range of further benefits.
Baydel has created a system with “virtual pathed” multiple RAID disk sets and all
other major components as foolproof, hot swap elements. This high availability
product offers outstanding performance and flexibility. In addition, the design offers
significant upgrade potential which requires little or no down-time. This further
improves the cost of ownership and avoids the disruption of periodic “fork-lift
changeouts.”
Engineered as a versatile storage “building block,” the O/ESR 2000 can be used on
its own in Direct Attach Server configurations, or in multiples within a storage pool
when configuring NAS and SAN environments.
With its current features and planned future upgrade path, the product is an ideal
solution for storage requirements ranging from a local application server to complete
enterprise-wide configurations.

292
The SAN Book III – October 22, 2001

Figure 279. Baydel’s O/ESR 2000

Source: Company website

➤ Ultra-SCSI modular RAID subsystem with failover, UPS, Battery Backup and
Caching. Capacities from 34 GB to 276 GB of usable disk space.
Figure 280. Baydel’s O/ESR RAIDER-5

Source: Company website

Baydel’s software development is dedicated to providing users with enhanced


features and functionality at the application level. RAIDMON, Baydel’s RAID
monitoring software, gathers performance data directly from the RAID controller.
RAIDMON shows system administrators exactly what their systems are doing and
enables fine-tuning, which leads to increased efficiency.
While most OSs provide no information about the internal condition of a RAID
system sitting on a SCSI port, an option is to write a device driver that can monitor
and control the internal RAID elements via the SCSI bus. A second option is to
keep the disk device driver standard (making life easier for installation and
maintenance), and to do all the monitoring and management via another port (e.g.,
RS232) from a user-level piece of software. This was the option Baydel chose —
with the advantage that the same Baydel RAID system can now connect to every
platform without concerns over device driver compatibility.

293
The SAN Book III – October 22, 2001

Figure 281. An Example RAIDMON Software Screen. Error Notification: Failed Disk

Source: Company website

Baydel works closely with value-added resellers that can specialize in particular
platform or application environments and provide high-quality integrated solutions
and services. Baydel offers support groups in Leatherhead and San Jose, direct
service in many North American cities, and trained service partners, including IBM,
SUN, Digital, Silicon Graphics, and Motorola in certain geographic territories.

BlueArc Corporation (Storage Subsystems)


Founded in 1998, Mountain View, California-headquartered, BlueArc has developed
a new high performance architecture for servers. The company’s SiliconServer
architecture provides an efficient solution for safe, easy to manage storage access
with solid state reliability and the scaling power to bring higher levels of server
performance at a lower cost.
BlueArc’s work focuses on the bottleneck in the flow of Web pages, video clips, and
other files around individual corporations and the Internet. To address this
bottleneck, BlueArc designed its Si7500 servers so that all data processing and
movement is done in the hardware. Through the use of reprogrammable gate flash
memory chips, the servers can be upgraded or expanded quickly and easily from a
download and since all processing is in the hardware, they can operate at multiples
of existing server speeds. The result of this architecture allows 2 Gbps throughput,
up to 250 terabytes of storage capacity, tens of thousands of simultaneously
connected users, and 99.999% solid state reliability.

294
The SAN Book III – October 22, 2001

Figure 282. BlueArc’s Si7500

Source: Company website

On May 7, 2001, BlueArc completed a $72 million round of funding, bringing total
funding since the company’s founding to over $100 million. This round will be used
to increase company growth. Investors from this round include Weston Presidio,
Celtic House International, Apax Partners, and Patricof & Co., Dell Ventures,
EUROQUBE, NY State Retirement Co., J.P. Morgan Partners, and Fort
Washington Capital.

Broadband Storage (SAN Appliance)


Broadband Storage is an early stage company developing next-generation, network
storage solutions that dramatically simplify enterprise data management. The
company’s solutions combine the ease-of-use and cost-efficiency of NAS, with the
scalability and performance of SANs. The Broadband Storage architecture
combines hardware acceleration with a proprietary distributed file system and a data
management system.
Founded in 2000, Broadband Storage is a privately held company that completed a
$15.45 million Series A round of funding in March, 2001. Headquartered in Irvine,
California, the Company boasts a dynamic team of highly skilled software and
hardware engineers who have a proven record of solving complex technical
challenges and successfully delivering innovative products to market.

Bus-Tech (Storage Components)


Bus-Tech’s businesses consist of intranet-to-mainframe connectivity solutions, high-
speed data movement products, and channel connectivity solutions for OEMs, all of
which provide a broad range of offerings to complement the role mainframes play in
corporate networks. Its channel and data movement technology solutions offer
means for customers to accelerate intranet deployment today while maintaining the
flexibility to adapt to the needs of tomorrow. These solutions enable users to
seamlessly access applications and data, which reside on the mainframe via standard
browser-based clients, without sacrificing performance, reliability, security, or
manageability.

295
The SAN Book III – October 22, 2001

A privately held corporation, Bus-Tech has grown from a single-product company in


1987 to one that manufactures and markets a full suite of powerful mainframe
connectivity and data movement products, sold both directly to end users and
through resellers and OEMs. While the company and its product line have grown,
the company’s original goal has not wavered: to provide its customers with data
center connectivity solutions.
Product line expansion has occurred in three main areas: intranet-to-host
connectivity; high-speed data movement systems; and high-performance,
programmable channel adapters. Bus-Tech’s business units reflect the industry’s
need for rapid communications between mainframes and servers and between the
Internet or intranets and mainframes, regardless of network protocol or topology.
Bus-Tech’s technology provides a migration path to upgrade to new network
technologies such as ATM and Gigabit Ethernet, while leveraging the advancements
being made in intranet software.
As Bus-Tech has grown, it has developed partnerships with intranet software
providers, IBM, Novell, and Microsoft. In 1997, Bus-Tech added marketing
relationships with three key vendors in the industry — network infrastructure leader
3Com, mainframe processor vendor Hitachi Data Systems, and host software vendor
Interlink Computer Systems. Bus-Tech’s scalable intranet-to-mainframe
connectivity solutions and high-speed data-movement products take advantage of
the mainframe as the central data system and because of the need to protect
investments in these systems over time.
Bus-Tech was the first to provide a channel-attach device that directly connected
local area networks to IBM mainframes — the Ethernet LAN Controller — Model 1
(ELC1). Bus-Tech, in close collaboration with IBM, developed the industry’s first
high-speed ESCON and parallel channel adapters specifically for use with IBM’s
eNetwork Communications Server for Windows NT software. Bus-Tech’s channel
adapter drivers are the only certified and include drives for IBM’s eNetwork
Communications Server for Windows NT.
Bus-Tech’s DataBlaster line of software, designed for high-speed data movement,
reduces the overall time it takes to do gigabyte-level data transfers, such as those
required in data warehouse and data mining applications.
The Original DataBlaster provides a fixed configuration of one Parallel (4.5 MBps)
channel to a SCSI-1 connection.

296
The SAN Book III – October 22, 2001

Figure 283. Bus-Tech’s Original DataBlaster

Source: Company website

The DataBlaster 2 can be configured to support up to two mainframe ESCON or


Parallel channels, and four Ultra Wide SCSI connections simultaneously.
Figure 284. Bus-Tech’s DataBlaster 2

Source: Company website

Figure 285. Bus-Tech’s NetShuttle

Source: Company website

The company’s NetShuttle line of products leverage Internet-derived network and


software technology to uniquely provide users with a “plug and play,” scalable
solution for access to mainframe applications, regardless of network protocols or
topology. Bus-Tech has added availability of Gigabit Ethernet in its NetShuttle
product line, making it possible for mainframe shops to move to a Gigabit Ethernet
backbone within their network on a cost-effective platform.

297
The SAN Book III – October 22, 2001

Figure 286. Bus-Tech’s NetShuttle Product Features

Source: Company website

The company also offers an Embedded Linux Controller (ELC) development


platform, designed to speed the development of network and storage appliances.
ELC combines a high-performance RISC processor, with an open PCI mezzanine
architecture.

Cereva Networks (Storage Subsystems)


Marlborough, Massachusetts–based startup Cereva Networks provides Internet
storage systems capable of accelerating content delivery, and simplifying
infrastructures for Internet data centers (IDCs) and service providers. Cereva was
launched in April, 1998, but only came out of stealth mode in mid-January, 2001.
Cereva Networks’ products include integrated network and switch capabilities tied
to common storage disk arrays, scalable up to 230 terabytes in first generation, with
bandwidth of 128 Gbps; scalable to 512 Gbps. The second-generation product will
have capacities just under 1 petabyte. The Cereva storage system uses Seagate and
IBM disk drives with Cereva-built proprietary arrays. Cereva initially plans to sell
its Internet storage system directly to service providers — ISPs, ASPs, SSPs, Web
hosting companies and large content providers — who maintain massive Internet
data centers.
Cereva’s Internet Storage solution enables large content providers to deal with
spikes in demand and scale their content infrastructure. This new type of storage
delivers the following attributes allowing service providers to offer storage as a
utility:
➤ Maximum volume pooling: enables the flexibility to handle spikes in demand
and quickly add new services and subscribers through rapid, non-disruptive
changes to storage configurations
➤ Constant availability: ensuring an “always-on” data tone
➤ Integral multilayered security: of all content and infrastructure

298
The SAN Book III – October 22, 2001

➤ Comprehensive standards-based systems management: for ease of use and


seamless integration with existing network operations center management tools
To date, Cereva has received approximately $86.4 million in financing and $20
million in credit from Comdisco Ventures, Intel, Global Crossing Ventures,
Goldman Sachs, Matrix Partners, North Bridge Venture Partners, Oak Investment
Partners, Sumitomo Corporation, and Worldview Technology Partners.

Chaparral Network Storage, Inc. (Storage Networking


and Components)
Chaparral Network Storage is a provider of Intelligent Storage Routers and external
RAID controllers for open systems network storage solutions. The Company’s
Intelligent Storage Routers facilitate the interconnection of SANs with existing
SCSI-based servers and storage systems. Their external RAID controllers distribute
data across multiple hard disk drives to increase data transfer speeds and deliver
fault tolerance.
The company’s Intelligent Storage Routers enable seamless bi-directional
connectivity between SCSI devices and Fibre Channel networks, allowing
companies to take advantage of the benefits of Fibre Channel technology while
protecting their investment in new and legacy SCSI storage devices. The routers
utilize their external RAID controller platform but are configured with different
embedded software for SCSI peripherals including tape drives, automated robotic
tape libraries, and optical storage devices. These routers meet the performance and
availability requirements of both departmental and enterprise server needs as well as
advanced applications for SANs.
Chaparral’s Intelligent Storage Routers also have the ability to offer high-speed
backup over Fibre Channel in connection with software anticipated to be released by
Legato and other companies in the near future. The product also provides for server-
free backup, which allows users to perform backup operations directly from disk to
tape without first copying data to the host computer.
Chaparral’s external RAID controllers offer fault-tolerance features for mission-
critical computing, such as high performance, redundancy, security, and protection.
These products target the external storage business in the open systems server
market as well as the SAN marketplace.
Chaparral is shipping its 2 Gbps storage routers that allow for easy attachment of
SCSI devices to Fibre Channel–enabled servers or SANs and provide the
performance, reliability, and connectivity required for data-centric applications such
as electronic commerce and digital video.

299
The SAN Book III – October 22, 2001

Figure 287. Chaparral Network Storage — FS2620: 2Gbps Fibre Channel-to-Ultra160 SCSI Storage
Router

Source: Company website

➤ G6322: Ultra 160-to-Ultra 160 SCSI: maximum performance board-based


RAID controllers, 4.25-inch x 9.0-inch, 180 MBps, 15,000 IOPS.
Figure 288. Chaparral Network Storage — G6322: Ultra 160-to-Ultra 160 SCSI

Source: Company website

The Company sells its products to OEMs, including Eurologic Systems, MicroNet
Technology, Inc., Qualstar Corporation, Quantum Corporation/ATL Products, Inc.,
Trimm Technologies, Inc., and Xyratex International Ltd., as well as to distribution
partners, including Arrow Electronics, Inc., Bell Microproducts, Inc., and CONSAN.
In late June, 2001, Chaparral announced the appointment of Mr. Victor Perez to the
position of chief executive officer. Mr. Perez has been Chaparral’s president and
COO since joining the company in December, 2000, and has served on Chaparral’s
board of directors since August, 2000. He will retain his president and COO roles.
Mr. Gary Allison, who had been Chaparral’s CEO since the company’s inception in
1998, will continue to serve as chairman of Chaparral’s board of directors.

CMD Technology, Inc. (Storage Components)


CMD Technology supplies peripheral storage controller technology and high-
performance storage-interconnect products to well-defined sales channels of OEMs,
VARs, integrators, and resellers. CMD Technology is privately held and employs
more than 160 people.
The company is headquartered in Irvine, California, and maintains sales offices in
California, Minnesota, Massachusetts, an Asian/Latin American sales office in San
Jose, California, and a European sales office in the United Kingdom. CMD has
strategic partnerships with companies such as Compaq (Digital), Intel, Microsoft,
Hewlett-Packard, Dell, and Gateway.

300
The SAN Book III – October 22, 2001

The company was originally founded in 1986 to develop and market SCSI host
adapters for Digital Equipment Corp. (DEC) computer systems. Today, CMD is a
leader in Fibre Channel and SCSI RAID controller solutions for UNIX, open
systems and PC marketplaces and is a leading supplier of IDE/Ultra ATA and
Universal Serial Bus (USB) semiconductors for the PC and embedded marketplaces.
CMD continues to invest in the computer peripheral interface and data-storage
technology markets
CMD provides data-storage technology for a wide range of open systems, high-
performance workstations, LANs, and PCs. CMD’s storage products are marketed
through system integrators, VARs and OEMs — many of the leading providers of
storage solutions. From CMD’s early success in the DEC market, many product
initiatives evolved. In 1991, CMD made the decision to add RAID to its product
line and to expand into the PC, UNIX and open-system marketplaces. Two years
later, CMD released the CRD-5000, its first RAID product.
The CMD Titan CRA-7280 RAID controller is a high-end SAN RAID controller
providing dual-redundant, hot swappable, active/active or active/passive RAID
controllers, power supplies, and cooling fans, dual Fibre Channel host interfaces,
and eight LVD Ultra2 disk channels. It provides true enterprise level storage
capacities and performance and comes bundled with an easy-to-use network-based
GUI RAID management utility.
Figure 289. CMD Technology — CMD Titan CRA-7280 RAID controller

Source: Company website

CMD also develops, manufactures, and sells high-performance ASIC and board-
level products. CMD’s IDE and Ultra ATA chips can be found in computer
systems, including those by ACER, AST, Compaq, Dell, Digital, Gateway, Hewlett-
Packard, and others. CMD also provides a wide range of complementary
software/firmware products and services that are designed to enhance the total
product solution.

CommVault Systems, Inc. (Storage Software)


CommVault Systems provides data management and robust storage management
solutions with the intent to meet the growing demand for reliable access to business-

301
The SAN Book III – October 22, 2001

critical corporate information. CommVault believes that major shifts in technology,


including the introduction of Windows 2000 OS, have opened significant
opportunities in both the traditional backup and SAN markets.
Formerly a strategic business unit of AT&T Network Systems, CommVault Systems
was founded in 1988 as an independent section of Bell Laboratories chartered to
develop an automatic back-up, archive, and recovery software product for AT&T’s
own internal use. Over the last decade, CommVault Systems has grown rapidly;
from commercializing its storage technology solutions in 1989 under Bell
Laboratories to becoming an independent, privately owned company.
CommVault Systems and Bell Laboratories’ pioneering efforts led to the
development of ABARS, an Automated Back-up, Archiving, and Recovery
Software product; a centralized storage solution in open networked computing
environments.
In late May, 2000, CommVault Systems released its Galaxy software suite, a storage
management solution architected around the logical rather than the physical view of
storage. Galaxy’s presentation of information provides granular access to stored
data, permitting improved restoration progress. CommVault Galaxy also improves
data and storage resource management, protection, access, and recovery. Designed
for Microsoft Windows 2000, the galaxy architecture works seamlessly with today’s
SAN and NAS systems. Galaxy is powered by an embedded Microsoft SQL 7
server meta database and enables the intelligence of storage management to be
embedded into SANs, storage devices, or Fibre Channel switches.
Galaxy delivers best of breed storage management solution for Microsoft Exchange,
Oracle, Red Hat Linux, UNIX, SQL server, NT server, Windows 2000 server,
clustering, NetWare, Lotus Notes R5/Domino, Network appliance systems, and
LAN-free backup with shared storage capabilities. CommVault Galaxy was recently
certified with Brocade’s Silkworm switches.
In June, 2001, CommVault announced the availability of Galaxy for EMC
CLARiiON NAS systems. The qualification of Galaxy software with CLARiiON
IP4700 systems offers customers a pre-tested, easy-to-use, data protection solution
that scales to handle multiple terabytes of capacity, delivering lower administration
cost, better access to application data and drastically improved backup and recovery
flexibility, according to the company. Earlier the same month, the company
announced that Galaxy is certified for 2000 Datacenter, Windows 2000 Advanced
Server, and Windows 2000 Server.
CommVault Systems is establishing an indirect distribution organization to
complement direct sales strategy. This indirect channel distribution will be made up
of international and regional integrators concentrating on NT applications. In
addition to the current international resellers Prosys, Acclaim Technology,
MicroAge, Bay Data, and Logical Choice, GE-IT signed an agreement with
CommVault.

302
The SAN Book III – October 22, 2001

CreekPath Systems (Storage Software)


Headquartered in Longmont, Colorado and formed by Exabyte Corporation in
January, 2000, CreekPath Systems creates solutions that enable Internet and
application service businesses to find, attract, and retain new customers, increase
overall customer satisfaction, and increase Web availability.
CreekPath Systems’ software is integrated with its storage management services,
and will be offered as a complete package to Internet data centers and managed
storage service providers in the near future. CreekPath’s technology will enable its
clients to deliver more efficient, fully automated storage network management and
online storage resources to their customers, at a more affordable price. Internet data
centers, application service providers, storage service providers, and corporate data
centers working as independent profit centers will use CreekPath technology to
monitor and manage their storage networks, to automatically service their
customers’ storage requirements, and to integrate the storage services into their total
service infrastructure.
CreekPath Systems has received $23 million in total financing. Key investors
include Exabyte, TeleSoft Partners, and Sequel Venture Partners. In July, 2001,
CreekPath Systems acquired Datavail Systems, which created end-to-end content
storage management and delivery system software.

Datacore Software (Storage Software)


Datacore Software operates on the belief that all storage resources should be
available over a network in a single pool, so storage can be easily distributed and
accessed exactly when and where it’s needed, no matter the hardware platforms,
operating systems, or storage devices in place.
Established in 1998, and an employer of more than 190 people, Datacore Software’s
primary product offering is SANsymphony, a program that converts islands of
dissimilar storage into a networked storage pool. This network storage pool makes
all storage on the network easily accessible. The software is scalable from the
workgroup level to full scale enterprise SAN and allows organizations to add and
redistribute capacity on the fly with a drag-and-drop interface. Administrators
allocate storage to individual computers with this interface, permitting volumes to be
shared between multiple cooperating servers by assigning appropriate read and write
privileges to different users.
Datacore Software designed SANsymphony for ease of use, allowing the integration
of new storage and servers more seamlessly. Changing computers is also simplified:
Network volumes need only be reconfigured from one system to another.

303
The SAN Book III – October 22, 2001

Figure 290. The Datacore Software Drag-and-Drop Interface

Source: Company website

SANsymphony is interoperable with the different components of a traditional SAN,


integrating UNIX, Windows, Netware, and Linux servers as well as diverse disk
subsystems, recognized hubs, and switches that form the Fibre Channel SAN
infrastructure. Providing interoperability with existing components, SANsymphony
allows for SCSI and SSA external storage disk arrays to be attached to the Storage
Domain servers and presented as Fibre-Channel resources from the network storage
pool using built-in protocol bridging.
SANsymphony also provides boosted retrieval times across high-speed SANs
because of the program’s caching technology. The introduction of Storage Domain
Servers boosts the usable cache along the data retrieval-path, minimizing direct disk
access and improving performance. Datacore also assigns multiple I/Os to the
physical storage devices, enabling a more efficient data stream, according to the
company. SANsymphony creates real virtualization of storage assets by making
virtual disk assignments, eliminating the costs and hassles of software
reconfiguration. The result is efficient, fast LAN-free backups that maximize
productivity and allow for current backup software, tape drives, and tape libraries to
be accessed on the SAN.
Datacore Software introduced three new interactive tools to help companies evaluate
the Return on Investment (ROI) and total cost of ownership (TCO) of storage
networking alternatives from different suppliers. The tools form the cornerstone for
DataCore’s expanded global professional services program, which helps customers
scope, design, deploy, and manage vendor-independent SAN solutions.

304
The SAN Book III – October 22, 2001

In June, 2001, JNI’s entire family of FibreStar Fibre Channel HBAs was verified
compatible and interoperable with DataCore’s SANsymphony software.
Additionally, the company announced that SANsymphony was selected as one of
the top five storage products of the year in InfoWorld’s, Reader’s Choice Award
issue.

DataDirect Networks (SAN Appliance and Storage


Subsystems)
DataDirect Networks was formed out of the merger between ImpactData, a SAN
software development company, and MegaDrive Systems, a provider of high-speed
Fibre Channel RAID-based storage solutions. Through the integration of these
companies’ hardware and software capabilities, the company has developed a
comprehensive SAN data system based on its DataDirect Network Architecture
(DDNA). The DDNA is composed of three main segments which can be sold and
implemented separately, or together as a complete SAN solution; the SAN Data
Director, an integrated SAN appliance, which incorporates switching, caching,
RAID and data management technologies; a SAN Operating System (SAN-OS)
which provides on-demand information access and a common interface for
heterogeneous systems and servers; and a SAN Data Manager, which provides
centralized management of the SAN network.
By providing intelligent data retrieval and management through its centralized
platform, DataDirect’s storage solutions are able to improve performance, increase
flexibility and lower the total cost of ownership with reduced equipment
requirements and low-latency shared access. The company’s SAN Data Director
delivers 1 Gbps performance per port even as additional devices and storage are
added to the SAN, helping to alleviate I/O bottlenecks and facilitating high-speed
data transfer among heterogeneous servers and storage systems.
Figure 291. DataDirect Networks’ SAN DataDirector

Source: Company website

The SAN DataDirector provides continuous access to your data and integrates easily
into multiplatform environments.

305
The SAN Book III – October 22, 2001

Figure 292. Multiplatform Environments

Source: Company website

Interoperability among these systems is enabled through the company’s DDNA


architecture, which includes its SAN-OS software that allows UNIX, Linux,
Windows NT, and Mac OS servers and workstations to use SAN data as if it were
locally attached. These systems can be incrementally expanded through the
interconnection of switches in high-density fabrics managed by the company’s Data
Director. Through its highly integrated approach to the SAN, DataDirect hopes to
evolve the SAN from a complex, unmanaged network of switches, routers, and
controllers to a transparent network of centrally-managed devices. This migration is
primarily enabled through the company’s advanced software platform.
The company also offers the SAN directION, which combines the full capabilities of
the SAN DataDirector SANappliance in a versatile “pay-as-you-go” model that
grows with users’ storage networking needs. SANdirectION allows users to build
an entry level SAN infrastructure with a minimum of two Fibre Channel host ports
(200MBps) and grow to eight (800MBps) host ports as needed. It is a network
infrastructure device, with intelligent front-end and back-end Fibre Channel
connectivity, that provides servers and workstations with instantaneous access to
shared storage. The SAN directION SAN appliance employs internal intelligence to
manage data access throughout the SAN infrastructure by creating host independent,
interoperable network storage pools by virtualizing host and storage connections.
This unique capability provides technology and vendor-independent storage pooling
as well as path and port independence.
DataDirect Networks offers SANware application suites, a series of SAN-certified
application suites that incorporate hardware, software, professional services, and
support to solve individual customer’s needs. DataDirect is able to leverage the

306
The SAN Book III – October 22, 2001

software base provided in its SAN Operating System and supported in its SAN Data
Manager to enable a centrally-managed control point from which its entire SAN
solution can be integrated. As such, DataDirect is able to significantly differentiate
its hardware offering and ensure it is positioned for future growth as new
capabilities, systems, capacity, and applications are added into its customer’s SANs.
DataDirect Networks also offers SAN RAID Systems, a family of products that
provide full redundancy, speed, high data availability, capacity, and scalability to
deliver terabytes of data in a single system from 200 MBps to 35 Gbps.
Figure 293. DataDirect Networks’ OEM Exclusive EV-5000, Fastest Fibre Channel Network RAID System
— 185MB/Sec Sustained Throughput

Source: Company website

Figure 294. DataDirect Networks’ EF-2000 Fibre Channel SAN Ready RAID Solution

Source: Company website

DataDirect Networks has sold over three petabytes of storage on more than 30,000
systems. The company’s customers include Amoco, Eastman Kodak, Microsoft,
NASA, Panasonic, Pinnacle Systems, Sony, and StorageTek. DataDirect Networks
has strategic partnerships with ATTO technologies, Brocade, Emulex, Gadzoox,
JNI, Qlogic, Seagate, and Vixel.

307
The SAN Book III – October 22, 2001

DataDirect Networks has introduced its “SAN Appliance Resource Management


Suite” for the company’s SAN appliance family of products. The SAN Appliance
Resource Management Suite enables administrators to centrally manage storage and
network resources, helping organizations reduce the overall costs of ownership and
increase the efficiency of their SANs.

Eurologic Systems (Storage Subsystems)


Founded in 1988, Eurologic Systems is a leading worldwide provider of network
storage technology to the OEM and indirect channel markets. Eurologic’s partners
include Dell Computer, Network Appliance, Lucent Technologies, Motorola, Agfa
Prepress, and Siemens Medical.
Eurologic is a global provider, with headquarters in Dublin, Ireland, and offices in
Asia, Europe, and North America. It has R&D centers located throughout Ireland,
the UK, and the United States, as well as an ISO-9002 certified, world-class
manufacturing facility in Ireland. It provides onsite product support around the
globe.
Eurologic Systems offers a comprehensive family of network storage solutions for a
wide range of markets. These products include the FLEXIbloc family of network
storage solutions based on Eurologic’s Flexible Storage Architecture, SANbloc
network storage system for storage area networks, Spheras Network Storage
Management software, the XL Series of fault-tolerant network storage solutions for
the OEM market, and the Voyager Series of systems for vertical market
applications.

ExaNet (Storage Subsystems)


Founded in 2000, and headquartered in Santa Clara, California, Exanet develops,
manufactures, and markets next-generation storage systems that enable massive,
seamless scalability of storage capacity and bandwidth, while maintaining ease and
efficiency of storage management.
Exanet has created an ultrascalable storage system, the ExaStore System. This
product scales in all dimensions, including capacity, bandwidth, backup/recovery,
manageability, geography and economy, while consolidating all storage devices into
a single, unified resource. The ExaStore System consists of patent-pending
ExaMesh software technology, running on a hardware platform built from standard,
cost-effective, off-the-shelf components. ExaMesh provides a fully scalable mesh
interconnect and a fully scalable storage cache. In addition to selling its Storage
Server directly to selected end-user customers, Exanet will license its software to
equipment manufacturers, who will create and sell their own ExaMesh-based storage
systems.
Exanet has a working prototype of its system now and is preparing for beta
deployment in late 2001. Exanet plans to ship its first Storage Servers in the first
half of 2002.
The company has so far raised $15 million in funding. Key investors include
Evergreen, Venture TDP, Keppel TatLee Bank, and Microdent.

308
The SAN Book III – October 22, 2001

Hitachi Data Systems (Storage Subsystems and


Software)
Hitachi Data Systems, a wholly owned subsidiary of Hitachi, Ltd, offers one of the
most extensible storage solutions available in the market today.
Key to the performance of the Lightning 9900 storage systems is the Hitachi Hi-Star
internally switched architecture. Hi-Star’s internal switching technology replaces
traditional shared-bus architectures. It helps to assure that high-speed data delivered
through the SAN fabric will not encounter a performance bottleneck inside the
storage unit itself.
Hitachi Data Systems (HDS) has successfully transitioned from a major mainframe
server player to a dominant supplier of storage solutions. Focused on becoming a
pure-play storage company, HDS has etched out a consistent track record of growth,
evidenced by its fiscal 2001 storage revenues of $1.2 billion, or 54% growth. The
goal: to become an $8 billion storage company by 2004 (which would represent
88% CAGR), through both organic growth and acquisitions. HDS has aggressively
attacked the enterprise (high-end) storage market by combining its leading edge Hi-
Star internally switched architecture with key partnerships such as Hewlett-Packard,
Sun Microsystems, and Veritas. We believe HDS is one of the most visible up-and-
comers in the storage industry.
A wholly owned subsidiary of Hitachi, Ltd (“HIT”), HDS offers one of the most
extensible storage solutions available in the market today to enable more scalable,
manageable, and cost-efficient data management. HDS’s Freedom Storage family
of products provides the foundation for Freedom Data Networks which leverages
SAN and NAS technology and supports multiple platforms including S/390, UNIX,
Linux, Windows NT/2000, and Novell NetWare.
In addition to its storage systems, HDS offers software and services to provide
storage on demand, remote copying for disaster recovery or backup, and
multiplatform data exchange. HDS also offers a full suite of customer-centric
services for data protection and SAN to provide individual configuration
management, system design, project management, installation, and support. To
facilitate implementation, HDS rigorously tests its products, as well as alliance
partners and third-party products, at its Interoperability Lab in Santa Clara,
California, as well as in its various R&D labs in Odawara, Japan.
The HDS Freedom Storage 9000 family of products comprises the high-end
enterprise class Lightning 9900 Series and the midrange Thunder 9200 storage
subsystems. HDS also offers its Freedom NAS product (a Network Storage
Solutions file server connected to either its Thunder 9200 or Lightning 9900 storage
subsystems), which enables customers to form a common storage pool for SAN,
NAS, and direct attach storage, in contrast to SAN-only or NAS-only offerings.
Management of and functionality for these systems are provided through Hitachi
Freedom Storage Management software including Hi Command, HRC (remote
copy), ShadowImage, and other software products. Additionally, HDS offers Just In
Time Storage, a capacity on demand program. Just In Time Storage is designed to
simplify the storage management challenges faced by large enterprises and Storage
Service Providers (SSPs).

309
The SAN Book III – October 22, 2001

Figure 295. Hitachi Data Systems Solutions

Thunder
Lightning 9200
9900

Source: Company website

Lightning 9900 Series (High-end Disk Storage Subsystem)


The HDS flagship product, the Lightning 9900 Series is available in Models 9910 or
9960 and ranges in capacity from 3.5 TB to 37 TB. These systems offer a 100%
data availability guarantee, with no single point of failure, hot-swappable
components, and duplexed write cache with battery backup. The Lightning 9900
offers one of the industry’s only internally switched fabric architecture, to deliver up
to 6.4 GBps internal system bandwidth. Operating through Hitachi Freedom
Software solutions, the Lightning 9900 is interoperable with UNIX, Windows 2000,
Windows NT, LINUX, Novell Netware, and S/390 hosts. The Lightning 9900
competes in the enterprise (high-end) market.
Thunder 9200 (Mid-range Disk Storage Subsystem)
Recently introduced to address the mid-range segment of the market, the Thunder
9200 is scalable up to 7.2 TB enabling responsiveness to unexpected user demand
and customer transaction surges. Consistent with HDS’s focus on reliability and
availability, the 9200 offers remote monitoring and an array of redundant and hot-
swappable components. The Thunder 9200 allows customers to establish
centralized, manageable storage control from a single location. This satisfies open
systems demands heterogeneously across a broad spectrum of environments,
including clustered Windows NT and 2000, Novell NetWare, and UNIX servers.
Key to the performance of the Lightning 9900 storage systems is the Hitachi Hi-Star
internally switched architecture. Hi-Star’s internal switching technology replaces a
traditional shared-bus architecture to provide the high levels of scalability required
to meet the growing information-access demands of e-commerce. The Hi-Star
architecture complements external switch connectivity to SANs and helps to assure
that high-speed data delivered through the SAN fabric will not encounter a
performance bottleneck inside the storage subsystem. The multiple, non-blocking

310
The SAN Book III – October 22, 2001

data paths created within the internally-switched architecture enable increased


throughput across workloads even as it is scaled to its full capacity (37 TB).
Figure 296. Hitachi Hi-Star Switched Architecture

Source: Company website

Key Alliances
VERITAS Software
On February 14, 2001, HDS, Hitachi, Ltd., and VERITAS Software announced the
formation of a Global Storage Management Agreement, which focuses on the
coordinated integration and testing of the companies’ storage solutions, in order to
architect highly interoperable data availability configurations that ensure increased
flexibility for the end user. As part of this Global Storage Management Agreement,
HDS will offer the latest versions of a wide range of VERITAS Software products
and professional services to its customers around the globe. The companies will
also unite in joint product integration and certification efforts.

Hewlett-Packard
On the OEM side, one of HDS’s top server partners is Hewlett-Packard. On April
28, 1999, HDS announced that under a three-year OEM joint technology agreement,
Hewlett-Packard will enhance its broad range of SAN and storage-management
products with HDS’s products and technology (this agreement was recently renewed
for another three years). More specifically, Hewlett-Packard will OEM high-end
subsystem technology from HDS (the Lightning 9900) and cooperatively enhance
products with Hewlett-Packard-contributed firmware and fibre channel interconnects
for improved open systems performance. Hewlett-Packard will also leverage HDS’s

311
The SAN Book III – October 22, 2001

manufacturing facilities and supply-chain management to optimize the availability


and flexibility of the Hewlett-Packard SureStore E Disk Array XP512, Hewlett-
Packard’s premier multiplatform high-end storage solution. Approximately 23% of
HDS’s revenues come from Hewlett-Packard and other OEM partners while
approximately 25% of sales are derived from its indirect channel partners such as
Datalink, GE Access, and Avcomm.

Sun Microsystems
On August 8, 2001, Sun and Hitachi Data Systems (HDS) announced a three-year
partnership agreement. Under the terms of the agreement, Sun will include HDS’s
high-end Lightning 9900 in its new Sun StorEdge 9900 series. In addition to the
subsystem itself, the Sun StorEdge 9900 also includes access to three jointly-staffed
support centers and certification in SunPlex and SunCluster environments. Other
terms of the agreement include cross-licensing of software and collaboration on
future software products.
We believe HDS is clearly focused on expanding its breadth of storage OEM and
reseller partnerships. Based on HDS’s success, we believe these partnerships could
make a lot of sense in the near term. We expect revenues will grow as a result of
these partnerships and HDS’s stated intention to grow channel sales twice as fast as
direct sales going forward. HDS expects indirect sales to account for 50% of total
sales within two years.

I-TECH Corp. (Storage Testing)


I-TECH Corp. serves the SCSI and Fibre Channel Test Equipment market, offering
high-performance testing solutions to peripheral manufacturers, system integrators,
and network suppliers. I-TECH’s product offerings address a range of testing
requirements, from emulation systems to passive bus analyzers, development
systems to production test systems, and field service.
I-TECH provides three interface-testing solutions developed to meet demands for
high-speed Fibre Channel analysis. Beyond operating rates of two Gigabits per
second, the products also provide multilevel and multi-user “intelligent analysis”
capabilities. I-Tech’s IFC-3 tester operates at data transfer rates of 1 Gbps. The
recently introduced IFC-4 tester transfers either 1 or 2 Gbps, and allows for creation
of low-level primitives. Both testers operate on the EAGLE universal software
architecture that also drives the company’s full line of SCSI testers.

312
The SAN Book III – October 22, 2001

Figure 297. I-TECH — IFC-4 Fibre Channel Tester

Source: Company website

The company’s Satellite Fibre Channel analyzer offering includes the portable IFC-
30 and IFC-40 Fibre Channel analyzers that monitor two channels of data at transfer
rates of 1 or 2 Gbps, respectively. It also includes the IFC-3016 and IFC-4016
desktop models that simultaneously monitor up to 16 channels at the same
respective speeds — for optimum efficiency and control at an extremely low cost
per channel.
Figure 298. I-TECH —Satellite IFC-4016 Fibre Channel Analyzer

Source: Company website

Ikadega (Storage Subsystems)


Founded in 2000, and headquartered in Northbrook, Illinois, Ikadega is a company
that develops high-performance, low-cost hardware solutions for efficiently storing
and delivering rich media content and other large data files over the Internet and in
other IP-based networks.
Ikadega’s patent-pending DirectPath Architecture is a switched-fabric server
architecture specifically designed for highly optimized storage and delivery.
DirectPath is specifically designed to serve as the underlying architecture for
optimized and affordable storage and delivery devices. These devices form the basis
of high-performance systems and solutions spanning many markets, including

313
The SAN Book III – October 22, 2001

Internet-based streaming media, enterprise information services, and residential


video-on-demand. Direct Path is able to achieve storage capacity and delivery
performance in serving large data files at low storage costs through an
implementation of proven data transfer technologies that bypass bottlenecks in
traditional architectures.
Features include:
➤ high speeds that reduce the need for load balancing by utilizing internal
intelligent “load delegation” methodology;
➤ maximizing delivery and storage performance while minimizing administration
requirements, driving down the total cost of ownership;
➤ highly scalable, while maintaining a constant performance-to-cost ratio;
➤ high capacity and cost-effective storage functionality;
➤ reduced need to employ redundant resources.
Ikadega is creating its own open-standards DirectPath-based servers for direct and
OEM sales to large system manufacturers serving the targeted markets. The
company also plans to license its DirectPath Architecture to major OEM hardware
vendors and license DirectPath implementations to various niche market vendors.
Ikadega secured $8 million in its first institutional round of funding in March, 2001,
bringing the total funding received to $15 million. Key investors include Kettle
Partners LP II, BlueStar Ventures, Leo Capital Holdings, and OCA Venture
Partners.

InfiniCon Systems (InfiniBand Networking)


InfiniCon Systems was established in 2000, with the goal of addressing the market
opportunities associated with bandwidth and connectivity limitations in data center
infrastructure. The company is a premier developer of intelligent system area
network fabric and multi-protocol I/O infrastructure, enabled by InfiniBand
technology. The company’s products are designed to enable low-latency server-to-
server communication and eliminate the network and storage I/O bottlenecks
emerging in large data centers, resulting in higher bandwidth, better data availability
and lower administration costs. InfiniCon’s products will be introduced by the end
of 2001. InfiniCon Systems is headquartered in King of Prussia, Pennsylvania.

Infiniswitch Corporation (InfiniBand Networking)


InfiniSwitch Corporation, founded in the fall of 2000, is a developer of InfiniBand
switching solutions designed for the data center. The company has developed a
rack-dense InfiniBand switching solution. Its first generation 1u (1.75-inch) 32-port
InfiniBand switch, called Leaf Switch, is designed to provide high availability,
scalability, and multi-protocol flexibility at the core of the InfiniBand fabric.
InfiniSwitch technology will integrate with other InfiniBand products, reinforcing
interoperability, a cornerstone of the InfiniBand standard. In addition, its
upgradable designs include three generations of link speeds and should support

314
The SAN Book III – October 22, 2001

today’s existing network protocols to protect the customer’s investment.


InfiniSwitch is located in Westborough, Massachusetts.

InterSAN (Storage Software)


Headquartered in Scotts Valley, California, InterSAN develops highly scalable
applications-based Storage Area Management (SAM) software. The company’s
software enables enterprises to maximize the benefits of large, heterogeneous
storage networks.
Central to the company’s product is its Virtual Private DataPath (VPD) technology,
in which the InterSAN solution is based. VPD relates applications to their data and
everything in between, including multi-protocol switches, heterogeneous storage
subsystems, storage virtualizers, appliances, and multiple security implementations.
The solution provides improved application availability, data security, and service
level management. The InterSAN solution improves the operational efficiency and
business scalability of large enterprises by centralizing and automating management
of complex storage networking infrastructure. InterSAN products have graphical
user interfaces for ease of use and support all major platforms including Windows
NT, Windows 2000, Solaris, and Linux.
On July 26, 2001, the company announced a partnership with EMC to integrate its
software with EMC’s information storage systems. Other partners include Brocade,
Emulex, Hitachi Data Systems, JNI, LSI Logic, and Qlogic.
In its first round of financing in April, 2001, InterSAN raised $8 million. Key
investors in the company include Alliance Venture Management, Morgan Keegan &
Co., Wit SoundView Ventures, and Mr. Kumar Malavalli (Co-founder and vice
president of Technology, Brocade). The company has approximately 30 employees.

Lane15 Software (InfiniBand Software)


Overview
Lane15 Software, headquartered in Austin, Texas, is the leading developer of
management software for InfiniBand networks. Lane15 is well positioned in this
emerging technology with a highly experienced management team, strong financial
backing, and an active involvement in the InfiniBand Trade Association (“IBTA”)
that is determining industry standards. The first InfiniBand-based solutions are
scheduled for release in late 2001 and scale production is planned for 2002. We
believe that the company is establishing a leadership position in InfiniBand
management products and will play a key role as this technology is implemented in
the coming years.
InfiniBand is a developing technology that will replace the PCI I/O bus and
revolutionize the way data is transferred between servers and shared I/O systems.
This next-generation I/O architecture is poised to dramatically increase network
performance, reliability, and scalability. IDC Corp. estimates that more than 75% of
all servers shipped in 2004 will have InfiniBand connectivity. Testament to this
projection is the strong support from leading vendors such as Intel, Sun, Compaq,
Dell, Hewlett-Packard, and Microsoft.

315
The SAN Book III – October 22, 2001

Lane15 Software, named for the “virtual lane” within the InfiniBand architecture
that is reserved for management systems, is a developer of vendor-neutral fabric
management solutions that can be readily adopted by any InfiniBand technology
provider, including chip manufacturers, server vendors, networking device
providers, and I/O vendors. In early July, 2001, Lane15 successfully demonstrated
its Fabric Manager technology at the IBTA Spring Developers’ Conference. With
its strong backing and demonstrated product, Lane15 Software is actively advancing
this next generation switched fabric I/O architecture.

Managing the Infinite


Lane15 Software has developed a clear product strategy focusing on enhancing
networks’ reliability, availability, and scalability in an InfiniBand environment.
Lane15 Software’s first generation fabric manager (2001) will focus on management
of InfiniBand devices and subnets while the second generation manager (2002) will
target a heterogeneous, policy-based management. Lane15 Software plans to add
security and performance based features (2003) to its software and ultimately arrive
at a software solution capable of evaluating InfiniBand quality of service as well as
automated provisioning of resources. While the first and second generation
managers will utilize an OEM strategy, Lane15 plans to develop a channel strategy
for the distribution of future management products.
Currently in development for a 2001 rollout, Lane15 Software will deliver two
products: a Fabric Manager product to be distributed through switch and server
vendors, and a Management Agent to be distributed through all InfiniBand vendors.
The initial Lane15 Software Management Suite will provide the necessary
foundation for added-value management products in the future.
The comprehensive Lane15 Management Agent and Lane15 Fabric Manager
offerings will include:
➤ subnet management and general services management agents for servers,
switches, routers, and I/O devices;
➤ comprehensive fabric manager solutions, including redundant subnet
management options that can reside in servers, switches, routers, and I/O
devices;
➤ a management console for network-wide InfiniBand fabric management and
administration; and the
➤ ability to function with network products from any vendor.

316
The SAN Book III – October 22, 2001

Figure 299. Lane15 Software’s Product Architecture

Enterprise Directory
Console Console
Web GUI

http(s)
http(s) xml
xml
Lane15 Fabric Manager LDAP or
Active
Directory
Policy Engine
extensions
Mechanisms

Switch IOCTLR

TCA

HCA
extensions extensions extensions

Host Switch Target

Source: Company website

Lane15 Software views the adoption of InfiniBand to start in 2001 with InfiniBand
silicon available (see partnerships below), 2002 for initial production systems, 2003
for early mainstream adopters and large scale deployment in late 2003 to early 2004.
Banding Together
Lane15 Software and Banderacom, a fabless InfiniBand semiconductor company,
have formed a partnership to deliver integrated silicon, software, and development
tools to InfiniBand equipment manufacturers. Banderacom will provide the silicon,
and Lane15 will supply the software. Lane15 and Banderacom along with
WindRiver Systems, a provider of software and services for connected smart
devices, demonstrated a prototype of the industry’s first InfiniBand Ethernet router
target channel adapter at the Intel Developer Forum crossing another key milestone
in the development of InfiniBand products. Note: Lane15 is also working with other
InfiniBand silicon suppliers such as Mellanox and Intel.

Funding History
Lane15 Software was founded in April, 2000, by Alisa Nessler, a Venture Fellow at
AV Labs. Based in Austin, Texas, the company was incubated within AV Labs after
receiving initial funding of $750,000. Lane 15 has raised approximately $9 million
in two separate rounds from a broad range of investors including Intel, Compaq,
Dell, Austin Ventures, AV Labs, and Lightspeed Venture Partners.

LeftHand Networks (Storage Subsystems)


Located in Boulder, Colorado, LeftHand Networks is a provider of Network Unified
Storage (NUS) solutions. LeftHand Networks’ network-centric architecture lets
companies scale storage modules and increase storage flexibility.
Through its NUS approach, LeftHand Networks enables companies to scale storage
modules seamlessly, in either a block or file configuration. Unlike previous

317
The SAN Book III – October 22, 2001

approaches that rely on a single network connection to a storage volume, the NUS
architecture avoids CPU bottlenecks and increases flexibility by mounting modules
in parallel on the network. NUS uses the network itself as the storage bus which lets
customers choose how to organize their storage volumes across Ethernet or gigabit
Ethernet networks to emphasize performance, redundancy, or both. Striping and
mirroring are possible within, and among, LeftHand’s devices — regardless of
geographical distance. The overall performance is similar to that of a Fibre-Channel
SAN, on standard Ethernet networks. The NUS architecture combines the
performance of a SAN with the ease-of-use of NAS in a stand-alone, integrated
network storage module. The company is expected to begin product shipment in
third quarter 2001.
LeftHand Networks recently received follow-on funding for its Series A round,
bringing the total to $13 million. Key investors include Boulder Ventures, Sequel
Venture Partners, and Vista Ventures I.

ManagedStorage International (Storage Service)


ManagedStorage International (MSI) is a provider of online managed information
access and data protection targeting large enterprises and e-businesses. The
company was incubated inside StorageTek in 1997 and then launched in March,
2000. StorageTek retains a minority ownership in ManagedStorage.
MSI offers data storage services on a utility basis including storage-on-demand,
server backup, PC backup, and exclusive value-add software for content
management. These solutions reduce the complexity and risks associated with the
storage management activities necessary in high storage demand and Internet-based
business applications.
The company offers a suite of fully managed storage utility services, providing
virtually unlimited enterprise-class disk and tape storage capacity; centralized,
secure server backup and recovery; and fully managed information archives with
storage, backup, recovery, and lifecycle management capabilities. All services are
available on a pay-as-you-use basis and can be provided on an on- or off-site basis.
The company’s fully managed e-storage “Power” services, including
PowerCONTENT, PowerNODE, and PowerBAK Personal and Server Editions,
reduce the costs and complexities associated with data storage management and
enable companies to focus on expanding their core businesses.
Storage-on-Demand offers “pay-as-you-go, pay-as-you-grow” disk storage space in
a shared or dedicated disk infrastructure. ManagedStorage provides several different
service levels tailored to each customer’s unique needs. ManagedStorage sells
directly to the end client and wholesales the service for reselling. Storage-on-
Demand is available to users located in many Internet Data Centers. Private
offerings are also available for “Fortune 500.coms.”
ManagedStorage’s Content Management service is application oriented, consisting
of data repositories supplied and maintained by ManagedStorage and aimed at
vertical markets such as medical imaging. ManagedStorage’s services include
capturing, indexing, archiving, preserving, searching, retrieving, and publishing any

318
The SAN Book III – October 22, 2001

form of digital content, from e-mails to audio to still images and movies. We
believe application-oriented services aimed at vertical markets offer higher value-
add and likely command higher gross margins.
The company has raised over $60 million to date from investors including Great Hill
Equity Partners, Providence Equity Partners, First Union Corporation, J.P. Morgan,
and EMC. StorageTek still owns approximately 15% of the company.

Mellanox Technologies (InfiniBand Networking)


Mellanox Technologies is a private company providing solutions for the server,
communication, and data storage markets, based on the InfiniBand Trade
Association standard for mid- to high-end embedded systems. Founded in March,
1999, Mellanox is a semiconductor company that supplies I/O building blocks for
the Internet infrastructure. Design, engineering, quality, and reliability operations
are based in Israel, while business operations, sales, marketing, and customer service
are in Santa Clara, California. The company develops semiconductors for this
burgeoning Internet infrastructure in the server, communication, and data storage
markets.
In order to accelerate customers’ time to market, the company offers system product
development kits (PDKs) including functional systems, board design database, and
software drivers. Mellanox recognizes system vendors’ hardware and software, and
provides a mechanism to support legacy I/O controller cards while enabling a
smooth transition from traditional I/O infrastructure to the new generation of
InfiniBand fabrics. Its emphasis is on delivering robust, high-performance, scalable
devices to serve the Internet infrastructure I/O fabric market. The company works
closely with customers and partners to define products and develop the system level
understanding necessary in order to effectively serve these new markets.
Figure 300. Mellanox Technologies — Two-Port Card

Source: Mellanox

Mellanox Technologies products include its InfiniBridge family of devices


supporting the new InfiniBand architecture. The InfiniBridge family of products
includes Switches, Host Channel Adapters (HCAs), and Target Channel Adapters
(TCAs). Mellanox is shipping InfiniBridge silicon, development boards, and

319
The SAN Book III – October 22, 2001

software products to customers, marking the first commercial availability of devices


supporting both 2.5 Gbps (1x) and 10 Gbps (4x) links compatible with the
InfiniBand 1.0 specification.

NetConvergence (Storage Networking)


Headquartered in Santa Clara, California and founded in 1998, NetConvergence
designs and markets IP storage systems using the industry-standard iSCSI protocol.
NetConvergence develops IP storage solutions that leverage and extend existing
network technology. NetConvergence’s xeniphast iSCSI technology enables
established network and storage vendors to implement a full range of IP storage
solutions. NetConvergence is currently rolling out a comprehensive suite of
xeniphast iSCSI host drivers to its partners and will expand its xeniphast product
family to include TCP/IP-offloaded iSCSI network interface cards, storage interface
cards and intelligent storage routers. The xeniphast product family delivers high
performance and fully scalable iSCSI solutions for IP SANs.
The company’s underlying technology consists of the following software and
hardware technology solutions:
➤ patent pending switch architecture that eliminates unwanted protocol overhead,
thereby delivering up to 10x performance improvement;
➤ unique software architecture bypasses the need for custom hardware
development by using off-the-shelf hardware, resulting in a substantial reduction
in the time to market, high performance, and low cost of ownership;
➤ xeniphast software platform’s layered driver architecture can adopt any Network
Interface Card (NIC) technology including proven driver software;
➤ accelerated TCP/IP offload to ensure iSCSI will have performance comparable
to Fibre Channel. This involves using recently developed TCP/IP off-load
engines on NetConvergence’s hardware platforms;
➤ iSCSI, which is the premier open standard IP Storage protocol;
➤ NetConvergence’s hardware strategy involves using “state of the shelf”
hardware, including a new class of processor called Network Processing Units
(NPUs).
NetConvergence raised $5 million in its second round of financing.

Nishan Systems, Inc. (Storage Networking)


Nishan Systems is a data networking company developing end-to-end SAN
solutions through existing IP and Gigabit-based Ethernet. Nishan Systems’ solution
to networking storage is SoIP (Storage over Internet Protocol) which possesses
features of IP networks such as product compatibility, established standards, and
reliability as well as characteristics of SAN technology such as high availability and
performance. Nishan Systems believes its solutions are ideal for today’s data-
intensive businesses because they introduce interoperability to storage networking.
Nishan plans for SoIP to allow the connection of today’s storage interfaces, such as
SCSI, Fibre Channel, and InfiniBand, with the standard IP infrastructure.

320
The SAN Book III – October 22, 2001

SoIP has utilized the recent improvements in an IP infrastructure to provide vendor


interoperability for enterprise and service provider customer segments, as well as
opening the door to the newer online storage market. Recent developments in
higher speed Ethernet, Multi-protocol Label Switching (MPLS), and switching
fabrics are the basis of the technology employed by SoIP. Nishan believes its SoIP
offers the ideal framework to enable a single network for data, voice, and storage.
Nishan systems is a member of the SNIA and JIRO alliances.
Figure 301. Nishan Systems’ SoIP Product Family

Source: Company website

Nishan Systems has introduced three switches:


➤ The IPS (IP Storage) 1000 gateway includes two Gigabit Ethernet SAN ports
and two iSCSI/iFCP Gigabit Ethernet ports to extend IP storage fabrics across
MANs and WANs.
Figure 302. Nishan Systems’ IPS (IP Storage) 1000 Gateway

Source: Company website

➤ The IPS 2000 switch includes up to six ports: two Gigabit Ethernet connections
and four Ultra2 SCSI (80 MBps) or Wide Ultra SCSI (40 MBps) ports.

321
The SAN Book III – October 22, 2001

Figure 303. Nishan Systems’ IPS 2000 Switch

Source: Company website

➤ The IPS 3000 switch has eight MultiService Interface ports, enabling users to
mix Fibre Channel and Gigabit Ethernet connections.
Figure 304. Nishan Systems’ IPS 3000 Switch

Source: Company website

Nishan Systems also offers the SANvergence Management Suite, which is based on
SNMP and Java, for managing the SoIP devices,. The software enables users to
manage an IP storage fabric consisting of Nishan’s switches and gateways. The
graphical interface of these tools provides an intuitive display and modification of
configuration data — simplifying the setup of switches and gateways.
Figure 305. Nishan Systems’ SANvergence Management Suite

Source: Company website

Earlier this year, Nishan Systems formed a broad strategic agreement with JNI
Corporation covering joint research, development, and marketing opportunities for
the integration of Storage over Internet Protocol (SoIP), including iSCSI, solutions
with Fibre Channel–based SANs in Metropolitan Area and Wide Area Networks
(MANs and WANs). The companies have entered into a “first-to-market”
agreement under which the two companies will jointly market and sell their
products.
Recent news includes Veritas interoperability certification of Veritas Backup Exec
with Nishan’s IP Storage Switches as well as a joint announcement with IBM and
Intel that software for the IP Storage protocol, Internet Storage Name Service

322
The SAN Book III – October 22, 2001

(iSNS), has been released as open source code. In addition, Nishan announced that
Intel will also support the iSNS protocol by enabling the Intel PRO/1000 IP Storage
Adapter, which will be available later this year, to utilize iSNS information.

NSI Software (Storage Software)


NSI Software is a developer of data protection and availability solutions.
Headquartered in Hoboken, New Jersey and founded in 1991, the company was built
upon the goal to provide network administrators a solution to improve the reliability
and serviceability of their networks.
The company’s Double-Take product is a real-time data replication system for
Windows 2000/NT, Solaris, and NetWare that reduces downtime and data loss,
allows for centralized backup, and can be used to provide offsite disaster recovery.
Within two years of first marketing Double-Take, NSI licensed it to more than 5,000
companies. Additionally, Double-Take was awarded Network Magazine’s Product
of the Year for 2001.
Double-Take monitors changes continuously to open files as they occur. Only byte-
level changes are replicated to one or more servers over standard network
connections. Data is continuously protected, and an up-to-date copy is available in
an off-site location. Replication occurs continuously over any shared or private IP-
based LAN, WAN, or SAN connection. In the event of a disaster, Double-Take’s
failover capabilities allow a secondary server to stand in automatically for the
primary server, transparently.
Figure 306. NSI Software’s Double-Take

Source: Company website

323
The SAN Book III – October 22, 2001

NSI’s GeoCluster integrates with Microsoft Cluster Service to create geographically


distributed clusters with replicated data to eliminate the hardware dependency and
single point of failure of a shared disk cluster. Servers in an MSCS configuration
share one storage subsystem and must be in close proximity to each other — from a
few feet if connected via SCSI, to no more than several kilometers if connected by
Fibre Channel. In this arrangement, computers take turns accessing the shared
storage — a single point of failure that leaves the entire cluster vulnerable to
hardware-related or geographic disasters. With GeoCluster, each server has its own
data copy, and geographic limitations are removed.
All NSI products are developed for centralized cross-platform management and
interoperability and include patented data replication and failover technologies. NSI
also offers a line of network performance enhancement utilities including Balance
Suite for NIC I/O load balancing and fault tolerance.
NSI has strategic technical and marketing relationships with industry leaders
including Microsoft, IBM, Dell, Hewlett-Packard, Compaq, and Data General (a
division of EMC). NSI relies on its OEM partners and reseller channel to deliver
comprehensive solutions and support to business users.

nStor Corporation (Storage Subsystems and Software)


Headquartered in Lake Mary, Florida (near Orlando) and San Diego, California,
nStor Corporation develops information storage management solutions, including
enterprise RAID solutions, desktop storage systems, data storage enclosures, and
associated network management software. nStor Corporation develops its storage
solutions primarily for the PC LAN network and RISC-based UNIX
workstation/server markets.
The technology used as the foundation of the nStor RAID product family was first
introduced in 1994, when the Storage Systems division of Conner Peripherals began
investing in the CR6 RAID subsystem. Conner Storage Systems grew quickly and
was responsible for a number of developments in RAID subsystem technology,
including Single Connector Attach (SCA) drives for hot-swap capability, graphical
user interface (GUI) RAID alert management software and enclosure technology for
fault-tolerance and network manageability. nStor has since expanded its product
family to include a full range of disk array solutions, providing high availability
storage per subsystem, and support for a variety of operating systems, including
Novell NetWare, Microsoft NT Server, SCO UNIX, IBM OS/2 LAN Server, and
IBM OS/2 Warp. Its 1999 purchase of rival ANDATACO more than tripled nStor’s
size and will allow it to market its products over a much broader geographic base.
nStor’s product offering includes both SCSI RAID and JBOD products. The
company’s SCSI RAID product offering provides both high-capacity (eight-bay) and
high-performance (12-bay) solutions within the same 3.5-inch form factor. The
NexStor eight-bay SCSI High-Capacity Series products scale by 1.4 TB increments
and stores up to 4.3 TB. The NexStor 12-bay SCSI High-Performance Series
products provide twelve spindles per 3.5-inch for instantaneous data access. The
NexStor 3200S and 3250S high performance solutions are focused toward high-
availability MS Exchange server and database application environments, where the

324
The SAN Book III – October 22, 2001

number of drives (spindles) become crucial for immediate I/O access, according to
the company.
Figure 307. The NexStor 3250S — Dual RAID Controller SCSI Storage System

Source: Company website

The NexStor 802S and 1202S provide unique SCSI JBOD storage solutions that fit
in a compact space. The NexStor 802S accommodates up to eight 1.0-inch or 1.6-
inch Ultra160 SCSI disk drives, including the 180 GB drive, mounted horizontally
in a 3.5-inch (2U) enclosure, while the NexStor 1202S accommodates up to 12 one-
inch drives in the same form factor. NStor’s design supports 7,200 RPM, 10,000
RPM, and 15,000 RPM disk drives, and allows an easy upgrade path to an
active/active SCSI or Fibre Channel host-attach RAID solution.
Figure 308. The NexStor 1202S — JBOD SCSI Storage System

Source: Company website

nStor has ensured that all of its products are SAN-ready and has developed a number
of products specifically for the SAN market, including its GigaRAID/FC family
which offers Fibre Channel host connectivity with performance up to 200 MBps.
nStor sells its products through a worldwide network of OEMs, and distributors.
The nStor customer base ranges from small businesses to Fortune 1000 firms and
government organizations, including Motorola, Yellow Pages, Whirlpool, Bank of
America, Trans America, Union Pacific, Prudential Insurance, The Trane company,
Intergraph, Warner Brothers, EDS, Lucas Film Ltd., Dream Works, the U.S. Navy,
and the U.S. Air Force.
Additionally, nStor’s AdminiStor Plus RAID configuration and storage management
software suite is fully integrated for use with nStor storage solutions. AdminiStor’s
advanced capabilities give customers the ability to configure, monitor, and manage
their nStor storage resources via Web browsers on a LAN/SAN network or over the
Internet.

325
The SAN Book III – October 22, 2001

Figure 309. AdminiStor Storage Management Software

Source: Company website

Recent news includes an announcement that nStor had received an offer from a
private investor to purchase approximately 45% interest in the company for $12
million. According to the press release, the offer requires the company to cause all
of its preferred stock to be converted into common stock, as well as other conditions.
Also in the press release, management stated that the company will evaluate the
offer, confer with the holders of preferred stock and respond to the proposal on or
before a late July, 2001 deadline provided for in the offer.

OmegaBand, Inc. (InfiniBand Networking)


Based in Austin, Texas, OmegaBand was formed in 2000 to adopt the use of
InfiniBand technology to solve the SAN issues in today’s Internet infrastructure
topologies. The company is focused on developing products to meet high-
performance systems needs. OmegaBand uses its proprietary core architecture and
partners with third-party vendors to provide systems that solve I/O problems for
large-scale data centers, Internet service providers, application service providers,
telecommunications, and e-commerce companies.
OmegaBand provides solutions that allow data centers to independently scale their
network-based resources and direct the connectivity of the incoming “fat” data pipes
to all data centers. The company’s products support interprocess communication
and high-performance shared I/O in the data center. These products are designed to
reduce complexity and overhead, while seamlessly integrating into existing
infrastructure.
Figure 310. OmegaBand’s IBgate 1000

Source: Company website

The IBgate 1000 is the first of a new series of InfiniBand Architecture (IBA)-to-
Gigabit Ethernet (GbE) gateways available from OmegaBand. It provides a
standards-based IBA-GbE interface that combines performance with networking

326
The SAN Book III – October 22, 2001

flexibility, providing customers with a simple, cost-effective way to introduce


InfiniBand-enabled equipment to an Ethernet network.
Over the longer term, OmegaBand’s data center service platforms will allow
customers to scale independently for increased network traffic and let multiple
servers share high performance I/O. Future generation OmegaBand products will
expand significantly on the first generation, bringing value through use of
OmegaBand’s proprietary technology to greatly increase data throughput and to
provide a rich feature set. The company plans to introduce much faster, 4x products
early in 2002.
The company raised $9 million in a first round fundraising in September, 2000. Key
investors include TeleSoft Partners and InveStar Capital.

Panasas, Inc. (Storage Subsystems)


Headquartered in Fremont, California and founded in 1999, Panasas is a network
storage company creating a smart and scalable storage solution by combining
distributed storage software with low-cost, industry-standard hardware.
Panasas is focused on developing a better way for organizations to arrange storage in
a network by creating “smart drives” that work with distributed software that will be
fully compatible with existing storage products and offer improved data security,
performance, scalability, and management capabilities. The result will be gigabyte
throughput performance and a reduced total cost of ownership for storage systems.
With this system, networked data can be directly stored and retrieved by client PCs
instead of going through a bottlenecked server that is overwhelmed with data from a
variety of sources. The system is built on a foundation of industry standards
allowing customers to maximize their return on their current infrastructure
investment.
Panasas supports NFS and CIFS network file system protocols for UNIX and
Windows client access, NDMP protocol for backup, and Ethernet-based TCP/IP for
connectivity. Panasas is currently in the product development stage, with a product
launch expected by mid-2002.
Panasas received $10 million of financing in September, 2000.

Pirus (SAN Appliance)


Founded in 1999 and based in Acton, Massachusetts, Pirus is a developer of carrier-
class networking systems that connect Fibre Channel SANs with IP systems. The
company’s products will connect SANs with LANs and WANs to manage and
improve data communications. The Pirus solution converges IP and Fibre Channel
storage networking to build next generation networks that reduce storage ownership
costs and streamline data management through the development of a central
component that combines disparate infrastructures without displacing existing
protocols or storage systems.
Pirus’s Storage Utility Switch is designed to enable the creation of scalable, reliable,
highly manageable, and cost-effective storage networks. The Pirus infrastructure
incorporates multiple protocols (Fibre Channel and Ethernet) and data access

327
The SAN Book III – October 22, 2001

methods (block and file services), as well as supports a wide range of industry-
proven storage subsystems, under centralized and secure management.
Figure 311. The Pirus Storage Utility Switch

Source: Company website

The Pirus solution addresses key storage issues facing large enterprise customers by:
➤ unifying file (NAS) and traditional block (SAN) components, IP with Ethernet
and Fibre Channel;
➤ securely sharing storage network resources among diverse communities of users
through the Pirus Secure Virtual Storage Domains;
➤ scaling new connections, such as new servers or storage resources, and enabling
additional processing power or capacity on demand;
➤ ensuring integration with the installed base of existing equipment, including
Fibre Channel fabric and director-class switches;
➤ bridging and converging network environments to enable integrated
management, efficient network utilization and greater connectivity across the
LAN, WAN, and MAN.
As the evolution of storage networking is an ongoing process, Pirus seeks to
preserve existing investments in storage systems while providing a smooth
migration path to next-generation storage architectures.
In March, 2001, Pirus acquired Blue Spruce Networks, a storage software
engineering firm located in Wichita, Kansas. Financial details of the acquisition
were not disclosed. Blue Spruce Networks will contribute its expertise in the areas
of storage architecture design, fibre channel, storage software development, and OS
platform integration — which complement Pirus’s own storage and networking
experience. Blue Spruce Networks will remain in Wichita, Kansas, serving as a
Pirus branch office, and report directly to Pirus’s vice president, engineering.
Additionally, the same month, Pirus announced that it closed over $27 million
through its Series B financing round, bringing total funding to over $45 million.
Strategic investors in this round of funding include StorageNetworks and Veritas. In
September, 2001, Pirus received an additional $9.5 million in funding from JAFO

328
The SAN Book III – October 22, 2001

Ventures. The new funding is targeted for the continued expansion of the
company’s research and development, marketing, and sales efforts.
The company is currently involved in many industry standard initiatives such as
IETF IPS Group (IP Storage), SNIA IP Storage Forum, ANSI T11, and FCIA. To
date, Pirus has raised over $55 million from strategic investors including
StorageNetworks and Veritas Software as well as financial investors such as
BlueStream Ventures, Bessemer Venture Partners, Charles River Venture Partners,
Comdisco Ventures, GATX Ventures, JAFCO, Morgan Keegan, and Silicon Valley
Bank.

PolyServe (Storage Software)


Based in Berkeley, California and incorporated in 1999, PolyServe provides systems
software products that enable enterprise customers to more effectively share data on
SANs within their data centers. These products help customers greatly improve
manageability, scalability, and availability of server and networked storage
resources.
PolyServe provides infrastructure software that helps enterprise customers
implement manageable, highly available applications in their data centers based on
enterprise-class distributed computing technology. PolyServe’s SAN file system
product, the PolyServe Matrix Server, is a key component in implementing scalable
server farms using SAN storage. The PolyServe Matrix Server allows multiple
servers to share data on a SAN as a centrally managed, unified resource. The
product provides standard applications with concurrent high-performance read and
write access to data with a high degree of data integrity and recoverability.
PolyServe Matrix Server will be available in first quarter 2002.
Figure 312. PolyServe’s SANs Vision

Source: Company website

PolyServe’s current product, PolyServe Application Manager, allows data center


customers to implement highly available clustered applications. This product is used
worldwide to simplify implementation of clustered Web and application servers and
reduce associated system deployment and management costs.

329
The SAN Book III – October 22, 2001

Features include:
➤ comprehensive wellness management for high node count environments
➤ granular monitoring of server, blade, and application health
➤ any-to-any configurations with easy “Distributed Management Console”
interface
➤ continuous processing during multiple hardware or software failures
➤ starting, stopping, and moving applications around the entire server environment
➤ efficiency improvement of load balancing efforts
Figure 313. PolyServe Matrix Server

Source: Company website

PolyServe Matrix Server is server software that enables multiple servers to


concurrently read and write data stored on SAN. With PolyServe Matrix Server,
each server has direct concurrent access to shared data in the SAN.
Features include:
➤ concurrent, high-performance read and write access to shared data on a SAN
➤ highly scalable
➤ dynamic addition of servers and storage elements
➤ no single point of failure
➤ online recovery of file system without interruption
➤ single point of administration from any server
➤ no changes to applications required
Polyserve has received over $25 million in total financing. Key investors include
Greylock and New Enterprise Associates, as well as the founders of Storage
Networks.

330
The SAN Book III – October 22, 2001

PowerQuest Corporation (Storage Software)


PowerQuest Corporation develops software and technology that provide solutions to
complex storage management environments through data management, deployment,
and protection software. Focusing on the small- to medium-sized enterprise market,
PowerQuest’s current technology enables: server deployment and management,
hard-drive management, incremental real-time back-up, point-in-time replication,
sector-based disk imaging, data recovery, workstation imaging and deployment, and
data migration and elimination.
PowerQuest’s server deployment and management software is ServerMagic, a
software solution that supports both NetWare and Windows NT/2000. ServerMagic
allows IT professionals to deploy NetWare servers quickly, with a consistent
configuration or to upgrade NT or NetWare server storage by copying, moving, and
resizing volumes and partitions on the fly without destroying or omitting data.
ServerMagic 3.0, released in 1999, allows NetWare volume segments to be merged
into one easily-managed volume.
PowerQuest’s set of desktop solutions includes PartitionMagic, a storage solution
that allows computers to create, resize, and move hard-disk partitions without
destroying data. PartitionMagic resizes and moves the NTFS and FAT32 file
systems used by Microsoft operating systems. In terms of data protection,
PowerQuest recently released DriveImage 4.0, a program that periodically creates
and saves images of the hard drive, and Drive Copy, which allows users to safely
and easily copy one’s old hard drive to a new hard drive in minutes instead of hours,
according to the company.
Figure 314. Power Quest Desktop Solutions

Source: Company website

For enterprise product and service solutions, PowerQuest offers its PowerManage
Suite, which provides users a data availability solution for enterprise-wide Windows
NT and 2000 Servers. This suite allows users to quickly isolate and resolve storage
problems on a real-time basis. According to the company, the suite combines
PowerQuest VolumeManager, PowerExpert ST, and StorageMonitor, for a solution
that prevents storage problems before they occur, minimizes system downtime in the
event of a problem, and reduces the overall cost of network storage administration,
all while controlling critical data assets.

331
The SAN Book III – October 22, 2001

Figure 315. PowerManage Suite

Source: Company website

Prisa Networks (Storage Software)


Overview
Founded in 1994 and employing over 50 people, Prisa Networks is focused on the
critical area of Storage Resource Management within Fibre Channel SANs. Prisa
develops network management software, consisting of service level management
and performance management tools, for System Area Networks as well as SANs.
Prisa Networks is active in the industry as a member of the Fibre Channel Industry
Association and the InfiniBand Trade Association. With few other companies
focused exclusively on providing network management software for the I/O
network, Prisa is uniquely positioned to capitalize on the tremendous growth
expected for both System Area and Storage Area Networks.
Prisa’s suite of network management software, VisualSAN, emphasizes service level
management and performance management, supporting the requirements for high-
availability storage networking and network performance. VisualSAN helps system
administrators maximize network uptime, optimize performance, and reduce overall
cost of networked administration while improving productivity. The key to the
VisualSAN suites is that its software will operate on SANs built upon Fibre
Channel, InfiniBand, and high-speed Ethernet technologies.
Prisa’s flagship product line, VisualSAN Network Management Suite, is sold
through hardware and software OEMs and service providers, including server and
storage systems manufacturers, storage and data management software vendors, and
storage-intensive ISPs, ASPs, and SSPs. At the Intel Developer Forum Conference
in February, 2001, Prisa announced its plans to extend VisualSAN to support SANs
based on InfiniBand Architecture in the second half of 2001.
Seeing Is believing
Prisa’s Visual SAN Network Management Suite comprises two products:
VisualSAN Network Manager and VisualSAN Performance Manager. This suite
enables service level and centralized management of storage area networks,
generates a topology map providing updates of real-time link performance to allow
quick discovery of hot spots or high traffic areas, and produces the information in
intuitive graphs and displays that increase productivity.

332
The SAN Book III – October 22, 2001

VisualSAN Network Manager (NM)


VisualSAN NM is the first module of the VisualSAN software suite. It provides
service level management of Storage Area Networks through graphical
visualization, event correlation, and alert generation. Network Manager
automatically discovers, manages, and monitors multi-vendor SAN devices
including switches, hubs, routers, host bus adapters and storage systems, generating
a topology map that depicts the SAN network elements, servers, storage systems,
and their interconnects. This map, viewed via an intuitive GUI, indicates which
devices and interconnects are active and which need attention. Network Manager’s
Event Manager provides fault isolation, real-time event correlation and alert
notification, allowing service providers and enterprises to maintain the availability
of the SAN network.
Figure 316. Prisa’s VisualSAN Network Manager

Source: Prisa Networks

VisualSAN Performance Manager (PM)


VisualSAN PM is an add-on application within the VisualSAN suite that seamlessly
integrates with Visual SAN NM. Performance Manager monitors real-time
performance of SANs and renders this data in an intuitive visual format with alert
generation. Data collected by Performance Manager is used for historical and
trending analyses that can help predict when and where future problems may arise.
The program is designed as a modular portable software application that can be
easily integrated into OEM SAN appliance solutions, accelerating their time to
market.
In addition, Prisa offers VisualSAN Object Server that enables third-party SAN
management applications to take advantage of VisualSAN’s automatic discovery,

333
The SAN Book III – October 22, 2001

network topology association, and event management capabilities. Consistent with


NM, PM’s remote access, Web-enabled interface provides the flexibility to access
SAN performance data from anywhere at any time.
Figure 317. Prisa’s VisualSAN Performance Manager

Source: Prisa Networks

Prisa Networks is working with Compaq Computer to codevelop and bring to


market Open SAN management solutions. The development efforts will deliver
SAN management software for Compaq’s SANworks Management Appliance,
providing customers with enhanced management of mixed-storage environments
within an open SAN. Additionally, Prisa plans to work with VIEO, a private
company specializing in fabric management software in the InfiniBand space, to
extend VisualSAN to support System Area Networks based on InfiniBand
Architecture in the second half of 2001.
Funding History
Prisa Networks has raised approximately $20 million to date from investors such as
Compaq, GE Equity, and Intel Capital. In October, 2000, Prisa Networks entered
into an agreement with Compaq Computer to codevelop and bring to market Open
SAN management solutions. In the same month, the company received investment
funding of more than $11.8 million from Compaq Computer, GE Equity, Intel
Capital, and other private investors. The funding from these companies reinforces
their confidence in Prisa’s ability to become the leader in its industry space.

Raidtec Corporation (Storage Subsystems)


Founded in 1992, Raidtec is a leading designer of storage management software,
SAN RAID systems, and NAS products. Raidtec offers end users the ability to
manage their growing storage needs with a broad range of cost-effective storage
solutions. Raidtec has successfully brought a series of high performance RAID,

334
The SAN Book III – October 22, 2001

Fibre Channel, and NAS products to market, with several thousand sold in over 45
countries worldwide.
Raidtec provides the market with end-to-end Fibre Channel PCI RAID controllers
and systems. The company manufactures its products in company-owned facilities
in Atlanta, Georgia, and Cork, Ireland. The engineering team in Cork, focuses on
Fibre Channel, SAN software and RAID controller and enclosure design, and on
NAS technology through the nEngine business unit. Raidtec has developed a wide
range of SCSI and FC products covering Server attach, Network attach, and SAN.
Raidtec’s broad product offering enables the company to be a “one stop shop” for
distributors, integrators, and VAR’s RAID storage requirements. Raidtec’s value
proposition is that it delivers extra features, performance, and reliability that are on
the forefront of customer needs.
Figure 318. Raidtec’s Products

Source: Company website

Raidtec’s product strategy is based on four core technology areas: Fibre Channel for
high performance RAID and SAN, NAS systems with embedded RAID, Fibre
Channel and LVD SCSI subsystems, and RAID power and packaging from two to
120 bays per system.
Raidtec’s nEngine business unit is focused on the development of e-commerce and
Internet infrastructure solutions, providing embedded, industrial-strength server
appliances for online commerce to take place in a secure, robust manner. Raidtec is
focused on the NAS market with its software and hardware for the workgroup and
departmental NAS server appliance segment of the market.

RedSwitch (InfiniBand Networking)


RedSwitch is a fabless semiconductor company that develops and distributes switch
fabric silicon solutions implementing InfiniBand, RapidIO, and other industry
standards for the storage, server, networking, and telecom markets. Incorporated in
June, 2000, RedSwitch was formed by senior executives and design professionals
from HAL Computer Systems Inc. (a wholly owned subsidiary of Fujitsu Limited).
RedSwitch is focused on developing standards based switches, in addition to the
development of proprietary solutions customers.
RedSwitch has delivered two generations of switch fabric comprising the Synfinity
product family for OEM customers. This switch and interconnect technology is the
result of eight years of internal research in cc:NUMA server systems, physical I/O

335
The SAN Book III – October 22, 2001

technology, and high-speed clusters. These system interconnect solutions


demonstrated link speeds of 2.5 Gbps and fall-through latencies of less than 35
nanoseconds. Consequently, RedSwitch has already built switch fabrics with
bandwidths comparable to those of the two emerging industry interconnect
standards, InfiniBand and RapidIO.
RedSwitch’s existing core technologies, design methodology, technical team, and
operation offer the advantage of rapid time-to-market. RedSwitch will leverage its
existing experience in switch fabric by focusing on the deployment of future
products in the storage, enterprise servers, networking, telecommunications, and
other embedded systems markets.
With a unique vision on the design and management of the next generation of switch
fabric products, RedSwitch is focusing on three technologies for building its
switches and interconnects: InfiniBand, RapidIO, and Customer-specific designs.
Figure 319. Focus Technologies

Source: Company website

RedSwitch and Agilent Technologies have a joint development program focusing on


InfiniBand. Together, they unveiled a 160 Gbps throughput switch product for
InfiniBand architecture, the first product of their joint development efforts.
This eight-port InfiniBand switch, scheduled for sample availability from both
companies in the third quarter of 2001, is designed for a wide range of applications
including storage and data networks, servers and workstations, server clustering,
input/output adapters, and high-speed backplanes. The product is a comprehensive
solution that embeds serializer/deserializer (SerDes) technology on-chip for reduced
cost, increased performance, decreased design complexity, and faster time-to-
market. The switch supports 2.5 Gbps (1x) or 10 Gbps (4x) InfiniBand link speeds,
on each of the eight ports, for a maximum aggregate bandwidth of 160 Gbps (bi-
directional).

Rhapsody Networks (Storage Networking)


Headquartered in Fremont, California, Rhapsody Networks is a storage network
infrastructure company that develops products that will enable corporate enterprises
and managed service providers to realize the benefits of large-scale storage
consolidation.
While the company’s product is still in stealth mode, Rhapsody Networks’ mission
is to build a switch to sell to large enterprises and storage service providers.
Rhapsody hopes to focus on security and manageability for large operations.

336
The SAN Book III – October 22, 2001

Rhapsody has received an initial round of funding worth $12 million and is close to
completing a second round of funding expected to be around $40–$50 million. Key
investors include Accel Partners and Sequoia Capital.

SanCastle Technologies (Storage Networking)


SanCastle Technologies was established in May, 1998 and is a provider of SAN-to-
WAN and SAN-to-MAN switching solutions based on both IP and Fibre Channel
communications standards. The company is offering a full fabric switch that can
connect SAN “islands” over long distances using the IP and Ethernet standards.
Applications for the SAN/WAN switch include mirroring (asynchronously) as well
as remote data access and disaster recovery. Applications for the SAN/MAN switch
include synchronous remote mirroring and data replication.
SANcastle is offering an edge switch with routing capability. It is positioning itself
as a partner to the core switch vendors such as Brocade and McData, connecting at
the edge of the network with switching capabilities as well as the ability to translate
protocols. While the initial product will translate from Fibre Channel to IP,
translations to other protocols, such as iSCSI, InfiniBand, or ATM, are possible.
The company does not position itself as a switch company competing with Brocade,
nor as a router company simply conducting protocol translation.
SanCastle’s Switch, the SANStorm 3000, is a SAN Switch that offers complete non-
blocking class 1, 2, and 3 levels of service, high availability, and a scalable SAN
switching fabric. SanCastle’s SANView software uses SNMP to fully manage and
configure the switch. SANcastle’s offerings are a hybrid of the two functions,
combining switching and routing to provide long distance capabilities to a SAN
fabric. It will be possible to use the edge switch as the only switch in a SAN, but
this is not the company’s objective. Rather, SANcastle, with its eight-port offering,
is providing the ability to increase ports in the SAN fabric and an access point to
other networks.
SANcastle is targeting the MAN market in addition to the WAN market. MANs
based on Ethernet offer high-bandwidth, low-cost, easy-to-learn and easy-to-manage
alternatives to traditional TDM metro technologies such as SONET.
SANcastle’s GFS-8 allows a company’s FC and IP networks to work together
seamlessly as a single connected network fabric. GFS-8, with its OpenIP
architecture, provides bridging, routing, tunneling, broadcasting, and resource
allocation services across the FC to GbE boundary and back again. No changes are
required to either network since GFS-8 is compatible with both network standards.
GFS-8 uses enhanced Transmission Control Protocol to ensure delivery and delivery
sequence of FC packets over the IP network without sacrificing integrity.

Sanera Systems, Inc. (Storage Networking)


Sanera Systems, headquartered in Sunnyvale, California and founded in October,
2000, is developing a carrier-class data center switch that enables enterprise and
service providers to implement the next generation storage/system area network. The
company’s solutions are designed to enable service providers and enterprises to

337
The SAN Book III – October 22, 2001

centralize storage capacity and management, as well as, consolidate servers into a
large computing farm over all of the industry standard network interfaces.
Sanera Systems is currently in stealth mode and has yet to announce its official
product offering. According to the company, its disruptive, high-performance
hardware and software architecture design boasts unmatched switching capacity,
multi-protocol connectivity, and unmatched performance. This enables service
providers and enterprises to deploy and manage a very large distributed storage area
network and server farms in a low latency, high bandwidth, service rich fabric.
In November, 2000, Sanera Systems secured its first round of financing of $17.5
million. Participants included Storm Ventures, CMEA Ventures, E*TRADE
Venture Capital, together with certain private investors. In April and May, 2001, the
Company received an additional $10 million financing from Western Technology
Investments and other private investors.

SANgate Systems (SAN Appliance)


SANgate Systems, founded in January, 2000, is developing an Enterprise Storage
Appliance (ESA) that supports both open systems and mainframe storage
environments. By implementing the SANgate ESA, SANgate believes customers
can dramatically decrease I/O response times, lower the total cost of ownership, and
streamline their data management process. These products enable relationships
between any two storage subsystems, any vendor, and any model. By supporting
both open systems and mainframe environments with features such as remote
mirroring, point-in-time copies and data migration, this unique hardware/software
combination solves major interoperability issues found in today’s enterprise storage
environments.
SANgate is headquartered in Southborough, Massachusetts and has an additional
engineering team based in Even-Yehuda, Israel.
Figure 320. SANgate’s Enterprise Storage Appliance (ESA)

Source: Company website

338
The SAN Book III – October 22, 2001

Hardware
The SANgate Systems ESA is an enterprise-class storage appliance that supports
both mainframe and open systems environments and operates with many vendors’
storage subsystems (according to the company). This intelligent platform —
specifically designed to run data storage applications — lets an enterprise increase
the capacity of its storage subsystems while increasing performance. It delivers
massive amounts of processing power, all dedicated to data storage management. Its
open architecture and comprehensive APIs allow it to be adapted to support future
storage system standards.
Software
The SANgate ESA Storage Management Software is an enterprise-class software
that performs seamless remote mirroring between any storage subsystems from any
vendors. It protects information while increasing the performance of your existing
environment.
SANgate Storage Management Software runs on the Enterprise Storage Appliance
(ESA). In addition to remote mirroring, it provides point-in-time copying and data
migration capability. SANgate Storage Management Software is a vendor-
independent software for managing data storage that supports both mainframe and
open systems environments.

Sanrise Group (Storage Services)


Based in Dublin, California, Sanrise is a managed data storage service provider
targeting the Internet data center and enterprise markets. The Company’s business
model includes a “same-store” manufacturing and distribution model that provides
consistent replication of its “storagecels” (the means by which Sanrise delivers
managed storage services), which accelerate speed to market and time to revenue
while enhancing manageability and minimizing interoperability constraint. The
company’s infrastructure provides 99.999% availability and offers monitoring and
maintenance capabilities.
Storagetone solutions consist of instantaneous, as-needed access to terabytes of
“disk-on-demand” scalability via optical fiber; backup and restore of data; and
safestore vaulting services. Customers benefit from Sanrise’s flexible SAN solution
through greater network efficiency, complete automation and online monitoring and
management, thereby lowering their total cost of ownership. Other customer
benefits include data availability, reliability, security, scalability, and performance.
The Storagetone.com customer portal allows users to view real-time global news,
reports, and information about their storage.

339
The SAN Book III – October 22, 2001

Figure 321. Storagetone.com Portal

Source: Company website

The Storagetone Fibrecloud solution is a complete managed portfolio of data


infrastructure services including: backup and restore, disk-on-demand, safestore
(off-site archiving), and thestoragetone.com customer portal.
Figure 322. Storagetone Fibrecloud

Source: Company website

StoragetoneOS is the core of the company’s Global Storage Intelligence (GSI) in-
house operating system that enables storage resource management, monitoring, and
customer provisioning within each storagecel. The company has 45 storagecels that
provide managed storage services to over 600 customers across a global footprint
from Germany to Japan. Sanrise provides these customers with real-time views of
their managed storage services through its secure website, mystoragetone.com. This
enables charting of mission-critical information flows which allows an enterprise to
more effectively manage its business and forecast future storage requirements.

340
The SAN Book III – October 22, 2001

Sanrise also provides a full suite of professional services to solve its customers’ IT
data management challenges. By conducting a complete storage assessment, the
Sanrise team of storage services experts is able to determine a customer’s managed
storage requirements and design a customized managed storage solution.
Sanrise has partnerships with leading technology providers including Brocade,
Hitachi, Oracle, StorageTek, Sun Microsystems, and Veritas, in addition to Internet
hosting companies such as Exodus Communications. The company has raised over
$160 million from investors including Comdisco Ventures, Crosspoint Venture
Partners, Exodus Communications, GATX Ventures, Hitachi Data Systems, Hitachi
Limited, Morgan Stanley Dean Witter, Texas Pacific Group (ACON Venture
Partners), and Veritas.

SAN Valley Systems (Storage Networking)


SAN Valley Systems aims to deliver highly reliable, cost effective, and scalable
next-generation storage infrastructure solutions. The company develops high
performance end-to-end networking products that provide storage solutions for
enterprise, service provider, and carrier networks. The company believes that
extending SAN infrastructure across Metropolitan Area Networks through the use of
new Gigabit Ethernet technologies including faster and fatter pipes will address
growing storage demands and solve today’s IP performance issues. New
alternatives to traditional tele-networks such as dark fiber allow users to leverage
existing IP infrastructure.
SAN Valley’s IP-SAN connectivity solution, the SL1000 Gateway, is specifically
designed to enable customers to interconnect legacy Fibre Channel storage area
networks into next-generation carrier access networks at gigabit speeds. SAN
Valley NMS provides a single interface that enables complete management of the
IP-SAN network. The auto discovery feature automatically manages new SL1000s
as they are added to the network. Complete configuration management includes
detailed port and channel configuration, enabling administrators to match differing
requirements across the network. Multi-OS, Java-based programming allows easy
migration of SAN Valley NMS among administrators’ desktops — crucial in today’s
heterogeneous networks.
Figure 323. SL1000 Gateway

Source: Company website

In addition, the company also offers the SVS Management System, an integrated
management tool for comprehensive management of the SL1000 IP-SAN Gateway.
Administrators can manage network performance, run statistical analysis, and
configure the SL1000 through its graphical user interface.

341
The SAN Book III – October 22, 2001

The SVS Management System provides a single interface enabling complete


management of the IP-SAN network. The autodiscovery feature automatically
manages new SL1000s as they are added to the network. Complete configuration
management includes detailed port and channel configuration, allowing
administrators to match differing requirements across the network. Multi-OS, Java-
based programming allows migration of the SVS Management System among
administrators’ desktops.
Figure 324. SVS Management System

Source: Company website

SAN Valley has released its Aspen Fibre Channel Managed Hub to the marketplace,
offering the industry’s only lifetime warranty. SAN Valley is a member of the Fibre
Channel Industry Association and the Storage Network Industry Association.

Scale Eight (Storage Services)


Founded in 1999, San Francisco, California–based Scale Eight develops technology
to provide rich-media storage services. Scale Eight has built a global storage system
that handles thousands of terabytes of rich digital media and allows customers to
access rich digital media from anywhere across the globe and view all their files
through a single file system image.
Through its patented architecture and software, the company’s storage system can
harness the capacities of disk drives and processors over the Internet to manage
thousands of terabytes of digital media including graphics, animation, audio, and
video. Scale Eight operates an Internet storage infrastructure which is an intelligent
global network that stores, manages, and serves files over the Internet. Based on
patent-pending technologies, the Scale Eight MediaStore service provides customers
with a global infrastructure for storing and serving files, provided as a fully managed
service. Once stored, the information can be directly accessed via any authorized
server or end user web browser. Additionally, Scale Eight provides virtually
unlimited storage, available on demand, allowing customers to instantly expand
storage capacity without limits.

342
The SAN Book III – October 22, 2001

Scale Eight’s technology provides a fully managed service that simultaneously


delivers the following:
➤ Ubiquitous Access: The Scale Eight Global Internet File System automatically
replicates content across multiple Storage Centers, creating a unified view and
global access of the file system from any customer location around the world.
➤ Massive Scalability: Storage architecture scales to thousands of terabytes.
➤ Instant time to market: Capacity can be deployed and expanded within hours.
➤ Continuous availability: Uptime is guaranteed by service-level agreements and
24/7 service management and maintenance.
➤ Cost effectiveness: Economics that support Internet-based business models.
➤ Guaranteed reliability: Data is geographically mirrored across Scale Eight
Storage Centers to ensure protection against catastrophic events.
➤ Steady Operations: Scale Eight Storage Centers are overseen by the Scale Eight
Operations Center, which monitors and manages overall system operation
around the clock, every day of the year.
➤ Fully redundant architecture: provides no single point of failure.
Scale Eight’s MediaStore is seamlessly integrated with leading content delivery
networks and provides advanced capabilities for monitoring and controlling the
MediaStore service via Scale Eight ServiceView, Scale Eight’s Web-based
management platform.
Key to addressing enterprises’ security needs, Scale Eight’s product offering also
provides full data protection, with each file automatically replicated to a
geographically remote Scale Eight StorageCenter, carrier-class service, with
committed service availability of 99.99%, and 24x7 monitoring, management, and
phone support. The company’s Service Management Center serves as the central
monitoring and control facility for the worldwide MediaStore service. Scale Eight
operates four StorageCenters, including two in the United States and one each in
London and Tokyo.
Strategic partners include Akamai Technologies and Exodus Communications, while
financial investors include InterWest Partners, CenterPoint Ventures, Crown
Advisors, and Oak Investment Partners. The company has raised over $31 million
to date.

Seagate/XIOtech (Storage Subsystems)


XIOtech Corporation, a wholly owned subsidiary of Seagate, develops
comprehensive network storage solutions for business customers. Seagate acquired
XIOtech in December, 1999, and the acquisition united the leader in physical drive
technology with the pioneer in network storage to deliver enterprise-class storage
solutions to its customers.
XIOtech was established to design, manufacture, and market a revolutionary,
reliable, easy-to-use, efficient, and centralized SAN. XIOtech integrates its SAN

343
The SAN Book III – October 22, 2001

hardware and software into a storage solution that can become the foundation for
any organization’s storage utility. This convergence greatly reduces infrastructure
complexity, demands on IT resources and risk to data, while increasing data
accessibility. XIOtech products are optimized for dynamic enterprise storage
environments with compatibility for enterprise applications.
Figure 325. XIOtech Product Family

Source: Company website

The XIOtech MAGNITUDE was designed to incorporate all the components of a


SAN in one centralized, easy-to-manage, highly-available configuration. The
MAGNITUDE incorporates a high-performance RAID controller, an eight-port
Fibre Channel switch and logical volume management software, making it easy to
implement multiple SAN components from a wide variety of vendors. XIOtech’s
exclusive Real-time Data Intelligence (REDI) Software Family, which is based on
true storage virtualization, provides the foundation necessary to significantly scale a
SAN as business needs grow.
➤ The REDI Access Software Suite enables multiple servers to access the same
storage volume, allowing server clustering and automatic server failover.
➤ The REDI Control Software Suite allows configuration and monitoring of the
MAGNITUDE from remote locations as well as automation of many
MAGNITUDE processes.
➤ REDI Copy Software provides the ability to duplicate storage volumes. It
enables a zero window backup as well as data testing, versioning, and data
migration capabilities.
➤ REDI SAN Links and REDI SAN Links Replicator Software extend the
capabilities of the MAGNITUDE across an enterprise. Multiple
MAGNITUDEs can be connected to simplify capacity planning, provide remote
data protection, and offer tremendous scalability.
XIOtech is currently working with IP vendors to create certified IP storage solutions.
XIOtech has strategic alliances with: Brocade, Gadzoox Networks, Imation,
Microsoft, Novell, Oracle, and Veritas.

344
The SAN Book III – October 22, 2001

Shugart Technology (Storage Software)


Headquartered in Irvine, California, Shugart Technology is a developer of integrated
products for testing, monitoring, and analyzing a wide variety of data storage
systems.
Shugart Technology provides premier software tools that simplify, automate, and
speed SCSI device testing in demanding and diverse storage environments. The
company’s software tools ensure consistent and reliable storage device testing across
a broad range of operating systems including Windows NT, Windows 2000, Solaris
SPARC, Solaris x86, HP-UX, and Linux.
Shugart’s product offerings are focused on software tools that can be used to
enhance and simplify testing of SCSI-based storage devices such as tape, disk,
RAID, or storage library. Shugart’s suite of calc software tools includes
FIBREcalc+, FIBREcalc and SCSIcalc for the testing of SCSI on Fibre Channel and
SCSI based data storage systems. These products are designed to support
development, production testing, field service troubleshooting, fault isolation, and
end user product compliance testing.
Shugart’s software automates the most commonly applied protocols and test
applications, while also facilitating quick and easy custom test development.
Shugart’s software is also portable across all popular operating systems and
hardware platforms and applications can be quickly added to meet the requirements
of a rapidly evolving network storage market.
The foundation of the software suite is the calc software tools such as FIBREcalc+,
FIBREcalc, and SCSIcalc. These software tools include the following components:
➤ Buttons: Calc tools’ command buttons are used to execute SCSI commands,
sequences, scripts, or test applications specific to the unit under test. These
command buttons makes testing easy, convenient, and consistent. A set of
default command buttons and device-specific command buttons are provided.
➤ Sequences: Command buttons can be included in sequences for various testing
scenarios that require a series of ordered steps. Sequences make it easy to ensure
all required testing procedures have been successfully completed and can be
created with ease.
➤ Data buffers: The calc software tools include eight data buffers that provide
additional testing flexibility. These data buffers can be used for a variety of tests
such as functionality, performance, stress, and data integrity.
The calc software tools also include the following optional components:
➤ Scripts — custom programs that use a C-based programming language. Scripts
provide additional control over the testing environment because they can use
program functions such as variables and conditional statements.
➤ Application packages — fixed test applications that automate more complex
testing tasks. These test applications can be used for various testing scenarios
such as performance, functional, stress, data integrity, and interoperability tests
for specific devices.

345
The SAN Book III – October 22, 2001

➤ Remote management — allows users to use the calc software tools remotely.
Shugart Technology is currently looking for a first round of financing of $15–$20
million.

Storability (Storage Services)


Founded in 2000 and based in Southborough, Massachusetts, Storability is a storage
service provider that designs, installs and manages data storage for corporations with
large databases that are critical to the enterprise and Internet data centers.
Storability’s storage management services take on the storage burden for
organizations with critical data requirements in a wide variety of endeavors,
including medical, pharmaceutical, industrial, financial, and e-commerce. For these
large enterprises, Storability can build an open SAN from the ground up or take over
the company’s existing storage environment and manage either solution remotely or
on-site.
Storability’s solution leverages new technologies to address customer concerns for
scalability, security, control, flexibility, and return on investment. Storability’s
leverage comes through the use of remote monitoring and operations facilities
overseeing dozens or hundreds of individual customer storage systems from a few
central locations. Storability delivers value directly to customers through its Storage
Assurance Services family of offerings. Storability’s customers enjoy instant access
to the company’s storage expertise and best practices, best-of-breed technologies,
predictable expense, and high return — all while being assured a high level of
service and security.
Storability’s Storage Assurance Services condense the traditional process of
deploying storage infrastructure to two unique concepts in storage assurance
provisioning. AssuredStorage provides ongoing managed data storage, while
AssuredAbility provides access to Storability’s experience, knowledge, and
methodologies for storage solution development. Through these offerings,
Storability assumes responsibility for the design, deployment, and ongoing operation
of the customer’s storage infrastructure. Ongoing services include 24x7x365 remote
monitoring and management, a 24-hour help desk, and problem management
processes. In addition, each customer is assigned a designated AssuredStorage
Operations Center engineer who is knowledgeable with the customer’s IT
environment.

AssuredStorage — MSS
AssuredStorage provides storage capacity and services tailored to unique customer
needs on a “pay as you grow” basis. As with other managed storage service models,
companies do not need to purchase equipment. In addition, Storability assumes the
risks and burdens associated with storage, including implementation, product
reliability, storage management automation and oversight, monitoring and
equipment service, procurement, and capital financing.
AssuredAbility — Traditional Services
AssuredAbility are Storability’s consulting services for designing, implementing,
and managing open SANs. Storability’s personnel determine requirements and

346
The SAN Book III – October 22, 2001

develop deployment models, while highly trained technicians manage


implementation.
The company has raised over $18 million from Battery Ventures, Madison Dearborn
Partners, and Lightspeed Venture Partners.

StorageApps (SAN Appliance)


Formerly known as RAID Power Services Inc., StorageApps provides
comprehensive, customer-driven enterprise storage solutions. StorageApps provides
a host of open, flexible products that offer global companies complete and
independent choice of server platform (operating system), storage device
(hardware), and connectivity (SCSI, Fibre Channel, TCP/IP). StorageApps provides
a complete, enterprise storage infrastructure to blue chip and Fortune 100
companies. The company offers 24x7x365 worldwide customer service.
StorageApps offers its own SAN storage Server, SANSuite.
In addition, StorageApps provides a number of SAN applications allowing data
replication, data snapshots, storage virtualization, and data migration. StorageApps
is a member of the Storage Networking Industry Association and has formed a
strategic partnership with Dell.
At the core of the company’s appliance is SAN.OS, an operating system that enables
in-band virtualization with host and storage device independence. The SANSuite
portfolio also includes security, data replication, and point-in-time image software.
SANMaster is StorageApps’ application for Web-based device and topology
management. Dell’s PowerVault 530F includes StorageApps’s portfolio of storage
management software as well as the SAN.OS.
StorageApps SANLink is a SAN-in-a-box capable of bringing storage outside of the
box into the virtualization scheme. Its tight integration allows for performance
tuning and high availability configurations with n-way peers. The SANLink Plus
appliance unifies all NAS/SAN management into one centralized management
console and permits scaling of NAS across multiple vendors’ storage.
On July 25, 2001, Hewlett-Packard announced it would acquire StorageApps for
$350 million.

StoreAge Networking Technologies (SAN Appliance)


StoreAge Networking Technologies is a private, Israel-based enterprise storage
solutions provider. Founded in 1999, StoreAge’s solutions are aimed at open
system, host-independent, multi-server distributed enterprises. StoreAge produces
advanced SANs that are highly modular, scalable, and cost effective. These SANs
are possible through the use of StoreAge’s own hardware and software modules
such as PCI SAN controller, RAID Manager, Volume Manager, and other
proprietary applications. StoreAge offers data integrity and system availability
through redundancy of all system components and automatic failover techniques.
The company is a spin-off from IIS, Intelligent Information Systems Ltd, which has
transferred to StoreAge all its intellectual property rights related to storage and
storage networking technology, including the Distributed RAID technology.

347
The SAN Book III – October 22, 2001

StoreAge is a member of the Fibre Channel, Storage Networking Industry


Association, and Celestra Alliances.
StoreAge is also focusing on virtualization in enterprise storage and has teamed with
SanOne in an effort to bring this technology to market. The two-year agreement
calls for StoreAge’s SVM technology to serve as the core virtualization engine for
SanOne’s DAMstor modular storage system. DAMstor represents breakthrough
technology integration of the latest storage-on-demand and capacity utilization
software, multiprocessor-based RAID and SAN fabric connectivity providing price/
performance improvements in networked storage solutions. SanOne and StoreAge
will cooperate in developing additional innovative and vertically-focused networked
storage solutions.
Figure 326. Storage Virtualization Manager (SVM)

Source: Company website

Tantia Technologies (Storage Software)


Tantia Technologies, a wholly-owned subsidiary of BETA Systems Software AG,
provides enterprise backup and recovery solutions that leverage SAN technology.
The company is composed of enterprise SAN experts, who bring mainframe
technology to the open systems market delivering ultra-high availability, intelligent
SAN solutions. Tantia Technologies builds and markets the storage solutions
originally developed by HARBOR Systems, which was acquired by BETA Systems
in December 1998. Tantia’s product line provides integrated, enterprise-wide
storage-management solutions for heterogeneous IT storage environments.
Tantia Technologies’ software leverages the SAN implementation by separating the
control path from the data path. Using a storage device’s open system data
exchange feature as a transport mechanism, physical data moves internally through
the box directly over an ESCON Channel to the OS/390 tape or disk. Only control
information moves via the TCP/IP backbone. This results in a ten-fold increase in
performance, and an 80% reduction in CPU utilization at the host, according to the
company. Because data moves through the disk storage, the network remains
uncongested, avoiding typical bandwidth constraints associated with TCP/IP and

348
The SAN Book III – October 22, 2001

SNA. Tantia plans to support non-OS/390-centric SAN environments, with full


support for emerging fibre protocols.
Tantia’s product lineup includes:
➤ Tantia’s NSM Enterprise-class backup and recovery solution, which provides
centralized control of backups that are performed at various remote offices of an
organization or within individual departments at a single location.
Figure 327. Tantia NSM Enterprise-class Backup and Recovery Solution

Source: Company website

➤ Tantia Harbor Backup — an innovative, high-performance client/server solution


for backup/recovery, archive/retrieval, and automated disaster recovery for
distributed system environments, and allows for the re-centralization of all
backup data into the data center.
Figure 328. Tantia Harbor Backup

Source: Company website

➤ The Tantia Enterprise Agent Suite, which allows online backup and recovery of
business applications: Oracle, IBM DB2, Microsoft SQL Server, Sybase,
Informix, SAP R/3, Lotus Domino, and Microsoft Exchange.

349
The SAN Book III – October 22, 2001

Figure 329. Tantia Enterprise Agent Suite

Source: Company website

➤ Tantia High-Speed File Transfer (Tantia HFT) — high-speed file transfer using
a variety of high-speed connectivity options across enterprise platforms.
Figure 330. Tantia Technologies — RS/6000 Solution

Source: Company website

Figure 331. Tantia Technologies — Bus-Tech Datablaster 2 Solution

Source: Company website

Figure 332. Tantia Technologies — Crossroads PCI ESCON Solution

Source: Company website

350
The SAN Book III – October 22, 2001

➤ Storage Monitoring Manager (Tantia SMM) — graphics-based, real-time


control and management of tape silo clusters.
Figure 333. Storage Monitoring Manager (Tantia SMM)

Source: Company website

➤ Storage Optimization Manager (SOM) — BETA 55: Automated tape stacking


system ensures the most cost-effective usage of expensive storage resources.
Tantia is headquartered in Boulder, Colorado and also maintains a development
center in Calgary, Alberta. Tantia provides enterprise-class backup and recovery
solutions to Fortune 500 companies throughout the world.

Times N Systems (Storage Subsystems)


Headquartered in Austin, Texas and founded in 1999, Times N Systems is a provider
of next-generation midrange storage systems. The company’s parallel, share-
memory architecture is designed to significantly enhance the I/O performance and
storage capacity of IT infrastructures with flexible platforms that scale with end user
needs.
Times N’s parallel architecture takes advantage of multiple I/O capabilities for faster
data access, while enabling the system to scale efficiently. Times N Systems’
software tightly clusters parallel drives together into a cohesive solution, using
standard disk and networking technologies on a Windows or Linux platform. Times
N technology serves either files or data blocks, and runs applications while flexibly
integrating into existing infrastructure.
Times N Distributed Drives provide performance-oriented virtualized storage that is
not only fast, but also completely transparent to the operating system, end user, and
their applications. Multi-terabyte virtual hard drives are built from disks across the
system. These distributed hard drives take advantage of multiple I/O channels and
multiple disk heads for the fastest possible data access. RAIDO, RAID1, and
RAID1+0 capabilities are all provided as software features. The company’s unique
Distributed RAM Drives, ranging up to 24 gigabytes, consist of memory resources
contributed from individual servers within the system. Users access database index
files, website graphics, or other static data at RAM speed rather than disk speed.

351
The SAN Book III – October 22, 2001

TNS GUI-based configuration and monitoring tools make setup and administration
easy. The TNS 1600 solutions support Windows 2000 Server, Advanced Server,
and Redhat Linux, allowing easy migration to a familiar network platform,
increasing users’ comfort levels, speeding access times, and simplifying system
management. The company’s solutions use standard protocols and are compatible
with popular third-party backup and management tools.
The company’s product should appeal to a small or mid-sized business, where a
complex SAN or NAS has not yet been installed. Time N Systems’ solution could
provide an attractive alternative for companies not willing to install a rudimentary
Fibre Channel SAN configuration or high-end server.
The company has raised approximately $15 million in two rounds of fundraising.
Key investors include Austin Ventures, Convergent Investors, and Sternhill Partners.

Troika Networks, Inc. (Storage Networking)


Founded in 1996 and based in Westlake Village, California, Troika Networks is an
emerging multiprotocol controller storage networking company with expertise in
various protocols including SCSI-over-Fibre Channel, IP-over-Fibre Channel, and
VI-over-Fibre Channel. Troika’s product offering of intelligent controllers and
configuration management software provides essential elements for designing,
building, and managing Fibre Channel-based server clusters and storage networks.
We believe that Troika is well positioned to take advantage of the tremendous
growth in clustering and storage networking. Troika’s product offering addresses
one of the most critical issues facing IT administrators everywhere — the need for
simplified server cluster and storage network management with better utilization of
existing infrastructures. Additionally, Troika’s partnership with Network Appliance
and Hitachi Data Systems, and its involvement in the DAFS Collaborative, place the
company in a strong position to advance its new technology, which we expect will
promote network convergence to a single system area network.
Essentially, Troika Networks exploits data-center-wide SAN connections for all
server and storage networking by integrating the SAN, data center LAN, and cluster
networks into a single, multi-protocol network infrastructure based on underlying
Fibre Channel technology. Troika’s design makes it possible to avoid the practice of
adding separate network elements (for example, clusters of application servers, file
servers, and email servers and their associated protocols) as application requirements
demand them.

352
The SAN Book III – October 22, 2001

Figure 334. The Troika Networks Solution

Source: Company website

Troika Networks’ bundled Zentai Controller and SAN Command management


software allow administrators the ability to monitor and manage Fibre Channel
resources, as well as tune performance, security, and high availability, depending on
application requirements. Additionally, this bundle gives the network the ability to
recover from server-to-storage path failures within the MAN through its Path
Command software (included with the Zentai Controller), providing users with
automated and transparent path failover/failback capabilities. This continuous
availability of path failure tolerance increases SAN reliability. Dynamic Load
Balancing makes multiple paths available for re-routing data during failover; and
failback does not require human intervention. Using the “Set-and-Forget” interface,
users can apply Path Command to specific paths to maintain network stability. In
addition, Path Command is capable of self-management and will automatically
respond to changing network situations, making efficient use of storage resources.

353
The SAN Book III – October 22, 2001

Figure 335. Troika Networks’ Zentai Controller

Source: Company website

The PCI-based Zentai Controller provides an intelligent server portal to support a


storage and/or system area network. The Zentai Controller is a multifunction
controller with 1) SCSI for storage, 2) IP (Internet Protocol) for backup
management, and 3) VI (Virtual Interface) for high performance application sharing.
Controllers feature Troika-designed ASIC technology and Fibre Channel software.
Troika controller products provide policy features for Quality of Service (QOS) that
can be leveraged across multiple protocols. The Zentai Controller, compared to
other HBAs, NICs, and controllers, uses minimal host processing power, permitting
these cycles to return for application processes.
Included with the Zentai Controller is SAN Command. As mentioned above, SAN
Command is a software interface that allows users to monitor, manage, and
troubleshoot their Fibre Channel networks. Specifically, SAN Command software
provides a graphical interface for administrators to configure, monitor, and control
Zentai Controllers and the Fibre Channel network. Together, Zentai Controller
enables consolidation of a system network with a storage network, while SAN
Command enables monitoring and management of the network’s Fibre Channel
resources. Additionally, SAN Command allows administrators to tune performance,
availability, and security, based on application requirements.
Announced on April 9, 2001, Troika’s most functional software line is the SAN
Command Enterprise-Active Controller Manager (ACM), which provides active
configuration and centralized control of distributed controllers. ACM enables data
center administrators to actively manage Zentai controllers and gain visibility into

354
The SAN Book III – October 22, 2001

connected SAN devices from a single location anywhere on the network, and we
believe this type of technology is essential for large-scale networks.
As an enterprise-level device management solution, ACM serves as a stand-alone
application or can be fully integrated within higher-level management frameworks
to provide a comprehensive SAN management solution. Additionally, ACM allows
administrators to easily create highly available network configurations by centrally
applying Path Command to specific paths to provide dynamic load balancing and
automated failover and failback capabilities.
ACM improves SAN performance, enhances end user applications, and resolves
network problems by providing remote configuration and firmware and driver
upgrades, diagnostic and troubleshooting capabilities, and performance tuning tools.
This centralizes network control and helps to reduce network administration costs, in
our opinion, essentially freeing administrators from the manual installation of
software, controller setting configuration, and troubleshooting otherwise needed to
maintain an efficient Fibre Channel network.
To date, Troika Networks has raised over $150 million from a number of financial
and strategic investors. Early investors included Draper Fisher Jurvetson and
Dynafund Ventures. Later investors include New Enterprise Associates, Winward
Ventures, Amerindo Investment Advisors, Intel Capital, and Network Appliance.
Troika Networks has partnered with Network Appliance to integrate Troika’s Zentai
Controller technology into Network Appliance’s enterprise filters. Troika’s solution
utilizing the VI (Virtual Interface) standard will enable integration of Network
Appliance storage appliances within Fibre Channel storage networking
environments, powering high-performance server-to-server and filer-to-filer data
exchange. Troika is also a member of McData’s OpenReady program for
comprehensive SAN interoperability, and on February 27, 2001, Troika announced
that its Zentai Controller had passed interoperability testing for Brocade products in
the “Brocade Fabric Aware” program.
Troika is an active member of the Storage Networking Industry Association (SNIA),
the Direct Access File System (DAFS) Collaborative, the Fibre Channel Industry
Association, the Fibre Alliance, and the InfiniBand Trade Association.

TrueSAN Networks (SAN Appliance)


Founded in 1997 as Synapsys Digital, TrueSAN Networks is a privately held
integrator of end-to-end data-center class Fibre Channel RAID and Storage Area
Network solutions for enterprise networks and advanced computing applications.
These solutions are based on its TrueFibre architecture, a 64-bit RISC-based host-
independent RAID architecture that uses the 200 MBps Fibre Channel protocol
throughout the entire subsystem for maximum available bandwidth. Whereas all
other dual-loop implementations involve two 100 MBps loops that cannot share
throughput (according to the company), TrueSAN solutions offer an architecture that
delivers one continuous 200 MBps segment of bandwidth. With TrueFibre,
TrueSAN Networks can deliver nearly 200 MBps of performance at the disk level,
enabling significant data throughput efficiencies and performance increases.

355
The SAN Book III – October 22, 2001

In addition, TrueSAN Networks’ SANengage technology enables host-independent


deployment of the company’s SAN solutions. With SANengage, multiple hosts can
access a single solution regardless of platform or interconnect method. LUN
masking, access privileges controls, and volume locking capabilities are offered via
independent hardware. TrueSAN also employs Fibre Uptime technology, which
provides hot-spare support and dedicated LAN-free DLT backup so that multiple
disk and controller failures can occur without losing data or compromising
performance. In support of this offering, TrueSAN has developed its Enterprise
Care program, which features services such as 30-minute on-site response, storage
outsourcing, 24x7x365 toll-free support, and a single point of contact.
TrueSAN’s SANengine storage networking platform combines storage domain
management, adaptive caching, resource optimization, and other technologies to
empower open systems with carrier-class performance, manageability, and
availability. With the SANengines series, isolated islands of storage are unified,
bandwidth barriers are broken, and vendor-independent SAN management is
realized.
TrueSAN’s Paladin product is a storage architecture that combines a modular and
scalable storage foundation with intelligent storage network management software
and flexible connectivity options. The Paladin architecture consists of three primary
components: foundation, connectivity, and intelligence.
The foundation is based on a pool of high-performance fibre channel disk storage
modules with embedded hardware RAID functionality interconnected through the
MetaFabric subsystem. Storage modules are available in increments of 200 GB, 400
GB and 800 GB with 512 MB of cache and 2 Gbps of bandwidth per module.
The Paladin Storage Network System is scalable to 50 TB of storage capacity, 32
GB of cache, and 128 Gbps of bandwidth. With its modular design and embedded
fabric architecture, users can add storage on-the-fly without downtime, even to
existing volumes. In addition, Paladin incorporates storage virtualization, snapshot
copy, and active resource management as standard features.

356
The SAN Book III – October 22, 2001

Figure 336. Paladin C-series

Source: Company website

TrueSAN sells its products and services directly to end users and integrators in order
to leverage the advantages of the direct business model.

Viathan Corporation (Storage Software)


Headquartered in Seattle, Washington and founded in February, 1999, Viathan
Corporation is a storage infrastructure company developing revolutionary
technology and products to enable storage virtualization and load balancing of
electronic data.
Viathan develops cutting-edge technology to help companies manage the
exponential growth of data. Viathan’s unique storage virtualization software
provides a comprehensive data management platform for use by Fortune 1000
companies and e-commerce vendors with large and growing storage needs.
Viathan’s proprietary technology adapts to meet the evolving requirements of
storage environments by creating a single virtual storage system. Additional storage
devices can be aggregated seamlessly into the environment without interruption or
reprogramming of the clients. Once part of the virtual storage environment, files are
dynamically allocated across all available devices for efficient storage utilization.
Viathan’s core technology is a flexible, scalable infrastructure for virtualizing and
managing NAS devices. Architecturally, Viathan’s technology is a layer of
infrastructure that sits between clients and storage devices. The primary purpose of
this layer is to virtualize the data sources from the perspective of the client. Layered
on top of the base technology are “value-add” components enabling heterogeneous
storage devices to be efficiently deployed, accessed, diagnosed, and managed by
corporations or storage application service providers.
On October 2, 2001, Viathan was acquired by Dantz Development, a privately-held
company that develops backup software.

357
The SAN Book III – October 22, 2001

Vicom Systems (Storage Networking)


Vicom Systems is a leading supplier of SANs with more than 1,000 systems
installed worldwide. Based on innovative storage routers, Vicom’s SANs provide a
new generation of RAID storage systems and a cost/effective means to create Fibre
Channel storage fabrics.
Vicom’s Serial Loop IntraConnect (SLIC) technology provides seamless
connectivity of FC-AL, SSA, or SCSI cabling in copper or glass fiber with up to
5.12 GBps internal bandwidth. Its SLIC technology is offered in two pre-packed
product lines; a SLIC storage system with sizes ranging from 81 GB to 147 TB and
a SLIC TeraRouter FC which utilizes a more-efficient FC-AL architecture than
traditional methodologies of stacking hubs and switches.
The company’s products are shipped with Vicom Systems Storage management
software, called SLIC Manager, which incorporates features such as remote copying,
dynamic configuration, exception monitoring and reporting, and off-site storage
access.
Vicom’s storage products and systems are compatible with all NT and Unix
platforms, and are compliant with all peripheral connectivity standards, enabling the
Storage Area Networks to accommodate a mix of incompatible hosts and multiple
storage device standards. With a Vicom SAN, data center managers can migrate
upward to new, high-performance data storage architectures while improving the
utilization of existing storage equipment.
Vicom’s storage products and systems are installed in data centers worldwide for
enterprise network and mission-critical applications and are supported with onsite,
24-hour, same-day service through a joint-venture with IBM Global Services. The
company’s products are sold through value added resellers and in partnership with
major computer and peripheral manufacturers.
Vicom’s SAN Virtualization Engine (SVE) supports storage pools that can scale up
to 500 TB for disk-level virtualization of Fibre Channel, SCSI, or SSA storage for
UNIX and NT servers. Vicom has extended its expertise to storage virtualization in
open systems environments with its fabric-based SVE appliance. SVE is built on
Vicom’s SV Router. The SV Router provides data and command routing paths
among servers, storage subsystems, and network devices. The SVE also includes
software with a Web-based interface for centralized monitoring, configuration
management, copy services, and zone management. The SVE is used in
combination with switches and hubs to expand the capability of the fabric since it is
network agnostic. Operation of the SV Router can be customized with UNIX scripts
and through an open API. Scalability is accomplished by adding routers to increase
bandwidth and connectivity.

358
The SAN Book III – October 22, 2001

Figure 337. Vicom Systems’ Independent Distributed Routing

Source: Company website

Figure 338. Vicom Systems’ SV Routers

Source: Company website

Figure 339. Vicom Systems’ SV Bridges

Source: Company website

359
The SAN Book III – October 22, 2001

Vicom and StorageNetworks have entered into an engineering development alliance


under which Vicom and StorageNetworks will develop storage virtualization
technology for StorageNetworks’ Global Data Storage Network (GDSN). Vicom
also announced that StorageNetworks has made an equity investment in the
company.

VIEO, Inc. (InfiniBand Software)


Overview
VIEO, formerly Power Micro Research, was established in 1994 to develop high
performance I/O solutions for a range of customers including leading system
platform and interconnect vendors, the national laboratories and government
agencies. Earlier this year the company was renamed VIEO reflecting its new
mission to deliver InfiniBand fabric management and application software. The
VIEO engineering team members have on average over twenty years of experience
in developing high performance systems. Members of the team have designed and
implemented general purpose and embedded operating systems, cluster and
distributed systems, high speed I/O architectures and systems and network
management software.
The company owns intellectual property that gives it a head start in developing
software to enable tomorrow’s InfiniBand-exploitive applications. VIEO’s software
will be deployed across all components of InfiniBand fabrics — in servers, switches,
and I/O devices. Optional interfaces to conventional management applications will
be provided so that IT users do not have to add another management tool to their
already complex environments.

Tight-knit Lineup
VIEO provides end-to-end, modular software for InfiniBand management. This
greatly simplifies the software integration task for partners because the entire stack,
from chipset control software to the management interface, is available from a single
source. VIEO’s ongoing product support and maintenance allow partners to remain
focused on their core business competencies. VIEO’s current products include:
VIEO Fabric Manager, VIEO FabricView, and the VIEO Software Developer Kit.
VIEO Fabric Manager
VIEO Fabric Manager includes all fabric management functions required by the
InfiniBand architecture, plus extensions specifically for VIEO. The software
framework, on which the management applications are built, provides support for
multiple chipsets, operating systems, and execution environments. The main
components of Fabric Manager are 1) core services, 2) subnet management modules,
and 3) general services modules.

Core Services
Core services provide a software framework for writing InfiniBand transports such
as Virtual Interface, Sockets Direct, and the VIEO Management API. This
framework is independent of the specific InfiniBand chipset, operating system, and
execution environment and exists at all levels of an operating system stack. All
VIEO provided applications are written using this framework.

360
The SAN Book III – October 22, 2001

Figure 340. VIEO Fabric Manager: Core Services — WAN/MAN Solution

Agents Managers

Management API

Common Services
Core Services Channel O/S Services O/S Bypass
Abstraction Adaptation Hardware
Layer Layer Library

Microcode and Chipset Support Firmware

Source: Company website

Subnet Management Modules


These modules are responsible for configuring the fabric, incorporating changes that
occur after initialization, and maintaining the integrity of the fabric. In order to
accomplish these tasks, the Subnet Manager performs several sub-functions
including topology discovery, local and global identifier assignments and mapping,
subnet partitioning, and topology sweeping.
Unlike Ethernet networks where switches dynamically learn the configuration of the
network by traffic analysis, InfiniBand fabrics are statically configured at
initialization time and in response to fabric changes and external policy management
requests. Topology discovery is how the Subnet Manager discovers fabric elements
and interconnects. Local and global identifier assignments assign locations and
program the switches to correctly route to them and provide all necessary mapping.
In addition to initial connections, subsequent events can change the topology of an
initialized fabric (e.g., links may break, nodes can be added or removed, or a port
may fail). The Subnet Manager sweeps the topology regularly to review its state, to
identify changes and to change the topology to incorporate the changes.

General Services Modules


These modules provide numerous tools such as management of communication
ports, data extraction, baseboard management such as power control and
temperature sensor reporting, and I/O configuration management such as power
control and temperature sensor reporting and I/O configuration. These modules also
include Networking Services that map InfiniBand services into the traditional
services domain, Node Initialization which is code that enables nodes to initially
attach to, and/or boot over the InfiniBand fabric, API and sample code modules that
enable partners to develop their own Vendor-Specific Managers that use VIEO
software framework, as well as many other monitoring and data extraction services.
VIEO FabricView
The VIEO FabricView family of software modules enables data centers to manage
InfiniBand fabrics and components using existing systems, storage, and network
management products. Companies that have already invested capital and human

361
The SAN Book III – October 22, 2001

resources to acquire, customize, and deploy these products can continue to use their
familiar human interfaces and semantics, thereby greatly simplifying the InfiniBand
management task and reducing cost.
FabricView gives software developers a convenient and stable programming
interface to the VIEO Fabric Manager. This interface provides two-way
communications allowing InfiniBand fabric management information to be passed to
the external management application for display and monitoring, and for
management commands to be sent to fabric components by the external application.
The combination of VIEO Fabric Manager and VIEO FabricView provides OEM
partners with a fully supported, end-to-end InfiniBand management software stack
— from the chipset control software to the management console.
VIEO FabricView includes several modules including Fabric Executive, an
application that aggregates fabric management data, serving as a single source for
requests from external applications, an API Library, that contains a set of
development tools for software developers, VIEO FabricView Console, which
allows access to management facilities using a JAVA-enabled browser, and
translation modules that allow external SNMP messages to pass through the
InfiniBand fabric. As with the Fabric Manager, third-party applications may also be
developed by VIEO partners and can use the VIEO FabricView API Library to
access the VIEO Fabric Executive in the same way as VIEO developed applications.

Software Developer Kit


The Software Developer Kit includes a developer-only license for VIEO’s
InfiniBand software stack enabling the card to be used as a Host Channel Adapter,
Target Channel Adapter, or switch. The Software Developer Kit also offers VIEO’s
end-to-end InfiniBand software stack (VIEO Fabric Manager, as described above
and VIEO FabricView Advanced Management Modules, which enable out-of-band
management of InfiniBand subnets using other VIEO products such as VIEO Fabric
Executive, VIEO FabricView API Library, and VIEO FabricView Console).
Licensees of the Software Developer Kit receive software updates as they are
published. This ensures partners stay current with VIEO’s latest maintenance and
functional enhancements.

Funding History
On February 5, 2001, VIEO raised $20 million in Series A funding from investors
including TL Ventures, One Liberty Ventures, Audax Ventures Fund, and Eyes of
Texas Partners. The $20 million in capital will be used to grow VIEO’s software
development and support teams as well as marketing and sales efforts to secure
additional partnerships, and to expand the company’s infrastructure at its Austin and
Richardson, Texas, and St. Paul, Minnesota offices. Moreover, VIEO just recently
announced an additional $4 million in Series A funding from BMC Software and
Dell.
Bringing It All Together
VIEO is conducting pre-beta tests of its technology, with plans to deploy a full beta
test with a number of undisclosed companies later this year. This includes the VIEO

362
The SAN Book III – October 22, 2001

Fabric Manager, which is currently in development and is expected to be


commercially available in late 2001.

Voltaire (InfiniBand Networking)


Headquartered in Bedford, Massachusetts and with its R&D facilities in Herzelia,
Israel, Voltaire is developing connectivity solutions for the Infiniband market.
Voltaire has created a product called nVigor which is a TCP/ IP to InfiniBand router
that provides accelerated connectivity, reliability, availability and scalability. The
product in designed to free the communication bottleneck between Ethernet TCP/IP
traffic and the InfiniBand Fabric. Additionally, nVigor offers application-based
intelligent connectivity with cluster-based centralized management.
Investors include Concord Ventures, Pitango Venture Capital, SFKT, Tamir
Fishman Ventures, TechnoPlus Ventures, and The Challenge Fund (Etgar LP).

YottaYotta (Storage Subsystems)


YottaYotta is the developer of NetStorage (the company’s term for networked
storage) technology for yottabyte (1 trillion terabytes storage). The YottaYotta
NetStorage Cube is based on the company’s scalable architecture which uses
distributed computing and clustering, and has at its core the NetStorage Array
Controller. The NetStorage Cube meets upcoming NetStorage requirements of the
Wide Area Network (WAN) for Internet backbone providers; Portal Area Network
(PAN) for Internet-based services such as SSPs and ASPs; Metro Area Network
(MAN) for metro loop storage; Storage Area Network (SAN) for corporations; and
Broadcast Area Network (BAN) for interactive broadcast providers such as video-
on-demand.
YottaYotta brings together the management and engineering teams of Seek Storage
Systems, a Washington-based company specializing in high performance storage
and Myrias Computer Technologies, a parallel super computers company based in
Edmonton, Alberta. YottaYotta’s family of products, which includes NetStorage
Cube, enables distributed direct access/clustering without the severe bottlenecks in
server controllers which are present in discreet, independent RAID systems.
Figure 341. YottaYotta’s Scalable Solution

Ethernet
FibreChannel
iSCSI
Host Channel Fabric
NetStorage
Blades
Provide:
Scalable
Bandwidth
and I/O’S
FCAL Drive Interface
Yotta
Storage
Capacity

Source: Company website

363
The SAN Book III – October 22, 2001

YottaYotta acquired intellectual property from QLogic. The technology, known as


AdaptiveRAID will further strengthen YottaYotta’s core I/O technologies portfolio.
As part of the transaction, YottaYotta will assume the QLogic facility in Boulder,
Colorado and it will become the company’s second strategic development center.
Figure 342. YottaYotta’s WAN/MAN Solution
City 1 City 2, Site A

Host Channel Fabric Host Channel Fabric

FCAL Drive Interface FCAL Drive Interface

Optical Core / Metro Network

Host Channel Fabric Host Channel Fabric

FCAL Drive Interface FCAL Drive Interface

City 3 City 2, Site B

Source: Company website

Founded in 1999, YottaYotta is a privately held company based in Montreal with


R&D in Edmonton, Alberta and business development in Kirkland, Washington.
The Company has raised $35 million to date with a $25 million second round closed
in August, 2000. Investors include Davenport Capital Ventures, Optical Capital
Group, Grosvenor Venture Partners III, Morgan Keegan, and TechnoCap.

Zambeel (Storage Subsystems)


Zambeel is a developer of technology to create solutions and platforms for data
management and storage. The company’s technology will allow users to store,
access, and manage large quantities of data on the Internet through high-speed
networks; and the company hopes to build the foundation of data centers of the
future.
Zambeel counts Kleiner Perkins Caufield & Byers (KPCB), New Enterprise
Associates and Integral Capital Partners as financial backers. In a recent Red
Herring article, Vinod Khosla, general partner at KPCB, mentioned Zambeel as one
of several new bets on real-time computing. Additionally, Zambeel was named a
2001 Red Herring “Ten to Watch” company. Zambeel is currently in stealth mode.

Zantaz (Storage Services)


Zantaz is a provider of Internet-based services and technology that can capture,
store, and retrieve business-critical communications (including email and related
attachments), documents, and transactions. The company’s first service offering,
Digital Safe is compatible with major email and telecom systems. Applications
include electronic message archiving, management and rapid retrieval solutions for
regulated industries and the worldwide corporate enterprise market. Zantaz also
provides professional services to implement, support and enhance the Digital Safe
service. First funded in 1998 and located in Pleasanton, California, Zantaz is a
privately held company.

364
The SAN Book III – October 22, 2001

Zantaz’s Digital Safe has been in development since 1996 and was first introduced
in March, 1999. It is a trusted service for the real-time archiving of high volume
Internet-based email, documents, and transactions. It fully meets the requirements
of the SEC and IRS for companies having regulatory requirements for archiving.
Key benefits include:
➤ outsourced service, no special hardware or software required
➤ highly scalable, allows you to pay only for what is used
➤ instantaneous retrieval with Web-based search engine
➤ transparent to users, no behavioral change required
➤ highly redundant for complete enterprise safety
Digital Safe’s technology draws heavily from the experiences of its founders, who
previously owned and ran traditional paper archiving warehouses. It comprises
highly redundant, geographically diverse data centers that provide complete
enterprise safety. Its proprietary, patent-pending technology brings certified
archiving to companies with regulatory and/or best-business-practice archiving to
their eBusiness operations.
Zantaz has current business relationships with Ernst & Young, Hewlett-Packard,
Cisco (as a member of the Hosted Applications Initiative), Sun Microsystems (as an
“executive” member of Sun Microsystems’ “service provider” program), Critical
Path, and Pilot Network Services, Inc.
Zantaz targets any company for which the retention and retrieval of documents is
part of their long-term business and legal programs and practices. This includes
companies such as securities dealers, brokers, and investment managers, and
applications such as online consumer financial products, corporate e-mail, storage
and data management.
Zantaz has a strategic alliance with Critical Path, a provider of Internet messaging
infrastructure, to provide an Internet-based messaging and eBusiness Archiving
service to the financial industry and corporate markets. The strategic alliance
expands distribution of the Zantaz Digital Safe services to Critical Path’s new and
existing corporate customers who will benefit from long-term email archiving
services.

365
The SAN Book III – October 22, 2001

This page intentionally left blank.

366
The SAN Book III – October 22, 2001

Glossary

367
The SAN Book III – October 22, 2001

Glossary

100Base-T Fast Ethernet A 100 Mbps technology based on the Ethernet/CD network access method.

10BaseT The IEEE 802.3 specification for ethernet over unshielded twisted pair (UTP).

Agent In the client-server model, the part of the system that performs information preparation
and exchange on behalf of a client or server application. In SNMP, the word agent refers
to the managed system.

Alarm vs. Alert In the most precise usage, an alarm is an off-normal condition (or state) indicating a
performance or equipment degradation or failure. An alert is an indication of a change
(or status) from a normal to an alarm condition or vice versa. However, the term “alarm”
is often used where “alert” would be more accurate. For example, “alarm dialout”
actually occurs on changes from alarm to normal as well as from normal to alarm.
ANSI American National Standards Institute. Standards-setting body responsible for the Fibre
Channel standard.

Application Layer Layer 7 of the OSI Reference Model; implemented by various network applications
including file transfer, electronic mail, and terminal emulation.

Application Program Interface (API) Means of communication between programs to give one program transparent access to
another.
Arbitrated Loop (FC_AL) A Fibre Channel topology that provides a low-cost solution to attach multiple
communicating devices in a single loop. It can support up to 126 nodes and one switch-
attached node in a single loop.
Archive To erase files from primary storage (disk), after a copy has been stored on tape or other
secondary media. The intent is long term storage where the file contents remain
unchanged. This function is also referred to as “retirement” or “shelving.”

Areal Density Bits per square inch of disk surface. Obtained by multiplying the bit density (bits per
inch) by track density (tracks per inch).
Array Management Software The body of software that provides common control and management for a disk array.
Array Management Software most often executes in a disk controller or intelligent host
bus adapter, but may also execute in a host computer.

Asynchronous I/O Requests I/O requests which bear no relationship to each other in time. Asynchronous I/O
requests may overlap in time. Independent access RAID arrays are able to provide
concurrent execution of asynchronous I/O requests which overlap in time, and whose
specified data addresses map to different member disks.

Asynchronous Operations Operations which bear no relationship to each other in time. Asynchronous operations
may overlap in time. The concept of asynchronous I/) operations is central to the use of
independent access arrays in throughput-intensive applications.

368
The SAN Book III – October 22, 2001

Asynchronous Transfer Mode (ATM) (1) The CCITT standard for cell relay wherein information for multiple types of services
(voice, video, data) is conveyed in small, fixed-size cells. ATM is a connection oriented
technology used in both LAN and WAN environments. (2) A fast-packet switching
technology allowing free allocation of capacity to each channel. The-SONET
synchronous payload envelope is a variation of ATM. (3) ATM is an international ISDN
high speed, high-volume, packet switching transmission protocol standard. ATM
currently accommodates transmission speeds from 64 Kbps to 622 Mbps. (4) Permits
hundreds of megabits per second to move on-demand throughout LANs, WANs and
desktops. Uses fixed length cells. Highly scalable, making it suitable for many different
kinds of traffic, from data and voice to video.

Autoloader A device that automatically inserts cartridges into a tape drive. Automation provides
unattended operation and helps ensure that back-ups take place on schedule.
Availability The probability that a system is available at time. Averaged over time, this function
expresses the expected percentage of time a system is available to do useful work.

Backbone (1) The part of a network used as the primary path for transporting traffic between
network segments. (2) A high-speed line or series of connections that forms a major
pathway within a network.

Backup (1) A copy of computer data that is used to recreate data that has been lost, mislaid,
corrupted, or erased. (2) A duplicate copy of data placed in a separate, safe “place”--
electronic storage, on a tape, on a disk in a vault--to guard against total loss in the
event the original data somehow becomes inaccessible. Generally for short-term safety.
Backup File Consolidation The ability for the backup software to recreate a full backup every time an incremental
backup is performed.
Backup Server A computer and storage system which provides backup and restore services for the
network as a network resource.
Backup Version A file, directory, or file space that a user has backed-up, which resides in an backup
storage pool in data storage. Though there may be more than one backup version of a
file in the storage pool, only one is considered the active version.
Backup-Archive Client A program that runs on a workstation and provides a means for users to backup,
archive, restore, and retrieve files. Contrast with Administrative Client.

Bandwidth (1) Measure of the information capacity of a transmission channel. (2) The difference
between the highest and lowest frequencies of a band that can be passed by a
transmission medium without undue distortion, such as the AM band--535 to 1705
kilohertz. (3) The maximum amount of data that can be transmitted through a data
channel per unit time. Usually expressed in megabytes per second. Analog bandwidth
is the range of signal frequencies that can be transmitted by a communication channel
or network. (4) A term used to indicate the amount of transmission or processing
capacity possessed by a system or a specific location in a system (usually a network
system).

Bridge/Router A device that can provide the functions of a bridge, router, or both concurrently.
Bridge/router can route one or more protocols, such as TCP/IP and/or XNS, and bridge
all other traffic.

369
The SAN Book III – October 22, 2001

Browser (1) Term used to describe the client program for the World-Wide Web. Popular browsers
include Mosaic and Netscape. Sometime called “navigator.” (2) Software program that
retrieves, displays, and prints information and HTML documents from the World Wide
Web.
Buffer A portion of storage used to hold input or output data temporarily.
Cache Memory A portion of memory dedicated to collecting and holding related data until a processing
or a storage module is ready to process it. The intent is to improve the overall system
performance. Cache for a disk drive is usually implemented as fast semiconductor
memory. See Read Cache and Write Cache.
Caching File System This is a process that operates on top of the file system to improve performance, data
accessibility, and client storage capacity. Client files are continuously copied to a
designated server so that they are backed-up and always redundant. Inactive files are
removed from the client to improve performance. Client desktop and portable file
systems are maintained and kept synchronized with the server.

CD/Compact Disc An optical disc that stores up to 650MB of data or 74 minutes of audio. CDs can hold
text, photo, image, audio and video files.

CD-R/Compact Disc Recordable Recordable CD technology that allows a user to write to a disc one time.

CD-RW/Compact Disc ReWritable Rewritable CD technology that allows a user to write to the same disc many times. HP
CD-Re-Writable media can be written up to 1,000 times per sector per disc.

CD WORM Compact Disc Write-Once Read-Many.


Central Administration A feature of network or enterprise software that allows a distributed process to be
controlled from an administration console. Control can reside both locally and remotely
either at any point or multiple points in the enterprise.
Channel A point-to-point link, the main task of which is to transport data from one point to
another. An electrical path for the transfer of data and control information between a
disk and a disk controller.
Class 2 Service A method of communicating between N_Ports in which no connection is established.
Frames are acknowledged by the receiver. Frames are routed through the Fabric, and
each frame may take a different route. In-order delivery of frames is not guaranteed.
Uses both buffer-to-buffer flow and end-to-end flow control.
Class 3 Service A method of communicating between N_Ports similar to Class 2 service, except there is
no acknowledgment of received frames. This is the lowest service level. Frames are
routed through the Fabric as in Class 2, and in-order delivery is not guaranteed. Uses
only buffer-to-buffer flow control.
Class 1 Service A method of communicating between N_Ports in which a dedicated connection is
established between them. The ports are guaranteed the full bandwidth of the
connection and frames from other N_Ports may be blocked while the connection exists.
In-order delivery of frames is guaranteed. Uses end-to-end flow control only.

370
The SAN Book III – October 22, 2001

Client (1) A recipient of services in a client/server application. Clients can be workstations or


other servers. (2) The part of Staffware which Users interact with. (3) A software
program used to contact and obtain data from a “server” software program on another
computer--often across a great distance. Each “client” program is designed to work
specifically with one or more kinds of server programs, and each server requires a
specific kind of client program.
Client Platforms The computer platforms from which the file or media catalogue can be accessed and
specific requests made.

Client/Server (1) The relationship between machines in a communications network. The client is the
requesting machine, the server the supplying machine. Also used to describe the
information management relationship between software components in a processing
system. (2) A system architecture in which one or more programs (clients) request
computing or data services from another program (server).

Compression Reducing the size of a data set to lower the bandwidth or space required for
transmission or storage.
Concurrency The ability of backup software to allow a storage device to receive data from more than
one system at a time; interleaving the data streams.
Controller A physical module that interprets signals sent between the host processor and a
peripheral device.
DASD Direct Access Storage Device. A term used by IBM in reference to disk drives.
DAT/Digital Audio Tape A 4mm helical scan magnetic tape technology originally designed for music and later
adapted for backing up computer-created data. DAT drives use cartridges that conform
to the DDS standard for data storage.
Data Accessibility A term defining the system requirements for network storage as continuous access to
files and data, high performance and protection from data loss.
Data Availability An application’s ability to access correct data in a timely manner. Usually implies
continued data availability in the presence of a component failure.

Data Compression Reducing the amount of electronic “space” data takes up. Methods include replacing
blank spaces with a character count, or replacing redundant data with shorter stand-in
“codes.” No matter how data is compressed, it must be decompressed before it can be
used.
Data Reliability Expressed in Mean Time to Data Loss (MTDL), is the average length of time over which
data stored in a disk subsystem can be correctly retrieved.
Data Transfer Rate The rate at which data is transferred to or from the storage media. It is usually given in
thousands of bytes per second (kbytes/second).

371
The SAN Book III – October 22, 2001

Database Backup The software has the capability to backup open relational databases. This is achieved by
a continuous monitor and backup of disk writes (raw partition) the backup of log files
which are files containing the continuous changes to the open database, or API
(Application Program Interface) database backup that allows the use of the native
backup process for the database application. Both methods are analogous to an
incremental backup policy that calls for an initial full backup only once and incremental
backups thereafter at intervals set by the system administrator.
Database Structure The underlying database engine on which the file and media database runs.
Density Generally, recording density.
DDS/Digital Data Storage The most widely used storage standard. DDS is the computer specification for 4mm
DAT technology. This standard was codeveloped by Hewlett-Packard and Sony.

Direct Access Storage Device Any on-line data storage device. A disc, drive or CD-ROM player that can be addressed
(DASD) is a DASD.
Director (Fibre Channel Director) A superior class of switching device. The term is not widely used in the Open System
world, and was introduced by McData as an extension of the IBM ESCON Director.
Disaster Protection A set of rules and procedures that allow a computer site to be put back in operation
after a disaster has occurred. The concept of moving backups off-site constitute the
minimum basic precaution for disaster protection. The remote copy is used to recover
data when the local storage is unaccessible after the disaster.
Disk A non-volatile, randomly addressable, rewritable mass storage device. This definition
includes both rotating magnetic and optical disks and solid-stale disks, or non-volatile
electronic storage elements. It does not include specialized devices such as write-once-
read-many (WORM) optical disks, nor does it include so called RAM disks implemented
using software to control a dedicated portion of a host computer’s volatile random
access memory.
Disk Array A collection of disks from one or more commonly accessible disk subsystems,
combined with a body of Array Management Software. Array Management Software
controls the disks and presents them to the array’s operating environment as one or
more virtual disks.

Disk Mirroring A fault-tolerant technique that writes data simultaneously to two hard disks using the
same hard disk controller. The disks operate in tandem, constantly storing and updating
the same files. Mirroring alone does not ensure data protection. If both hard disks fail at
the same time, you will lose data.
Disk Striping A type of disk array mapping in which consecutive stripes of data are mapped round-
robin to consecutive array members. A striped array, also known as a RAID Level 0
array, provides high I/O performance at low cost, but provides lower net data reliability
than any of its member disks.
Disk Subsystem A collection of disks and the hardware required to connect them to one or more host
computers. The hardware may include an intelligent controller or the disks may attach
directly to a host computer’s I/O bus adapter.

372
The SAN Book III – October 22, 2001

Disk-Based Backup The process of sending backup data to hard disk storage rather than directly to tape.
The backup storage pool on disk is managed for fast restore capability. Background
copies are made to tape or to a remote vault.

Distributed Computing Environment An architecture of standard programming interfaces, conventions, and server function
(DCE) personalities (e.g., naming, distributed file system, remote procedure call) for
distributing applications transparently across networks of heterogeneous computers.
Promoted and controlled by the Open Software Foundation (OSF), a vendor consortium.
DLT/Digital Linear Tape Quantum’s linear magnetic tape technology, used in many high-performance, high-
capacity data backup solutions.
Duplexing A configuration in which each element of a system or subsystem is duplicated, in
addition to the disk drive, example a controller, a host adapter, a power supply. See
Mirroring.
DVD/Digital Versatile Disk (or A CD-sized optical disc with potentially up to 14 times the capacity of a CD (when both
Digital Video Disc) sides of the DVD are used). A single-sided DVD disc might store an entire movie or a full
set of reference books.
EIDE/Enhanced Integrated Device An extension to an IDE interface that supports different IDE device types, such as hard
(Drive) Electronics disk drives, CD-ROM drives and tape drives.
Encryption Data can be coded into an unreadable form and decoding is restricted by password,
user access, or other means. Some software allows decoding to be transparent to the
authorized user, while to the unauthorized user, data is unreadable.

End User Any customer of an interstate or foreign telecommunications service that is not a
carrier, except that a carrier other than a telephone company shall be deemed to be an
“end user” when such carrier uses a telecommunications service for administrative
purposes. It is also a person or entity that offers telecommunications services
exclusively as a reseller shall be deemed to be an “end user” if all resale transmission
offered by such reseller originate on the premises of such reseller.
Enterprise Data Management This is a class of software that builds on Network Data Management software, adding
Software the capabilities of interoperating and providing backup and restore services across
multiple, heterogeneous computing environments. Backup and restore is performed
between any combination of PC’s, UNIX, Midrange systems, and mainframes.

Enterprise Network A geographically dispersed network under the auspices of one organization.
Enterprise Storage Management A family or suite of related processes and applications which manage storage devices
as enterprise resources across heterogeneous environments. Storage management
applications range from asset management to hierarchical storage management.
Ethernet A 10 Mbps, 100 Mbps, and 1 Gbps baseband local area network protocol that allows
multiple stations to access the transmission medium at will without prior coordination. It
resolves contention by using carrier sense and multiple access with collision detection
(CSMAICD). Defined under the IEEE 802.3 protocol.
Exchange A group of sequences which share a unique identifier is called an exchange. All
sequences within a given exchange use the same protocol

373
The SAN Book III – October 22, 2001

F_Port Fabric port. A port on a fabric switch to which N_Ports may be directly connected. An
F_Port is uses the address identifier hex’FFFFFE’.
Fabric A Fibre Channel network consisting of multiple devices interconnected by one or more
switches that use Fibre Channel methodology to link nodes and route frames.
Fabric Switch In this category of switch, any port on any switch can provide (subject to bandwidth
availability) full speed access to any other port on the network. The network consists of
a fabric of linked switches.
Failure A detectable physical change to hardware. Failures may be repaired by the replacement
of a physical component.

Fast LAN Term given to two emerging standards; IEEE 802.3u (called Fast Ethernet) for Ethernet
operating at 100 Mbps over Cat-3 or 5 UTP, and IEEE 802.12 (called 100VG-AnyLAN) for
Ethernet or Token Ring operating at 100 Mbps over CAT-3/4/5 UTP, STP or optic fiber.
Fault tolerance The ability for a system to remain operational, sometimes in a degraded mode, even
though one or more elements of the system have failed. Fault tolerance is obtained
through the use of redundant components or modules.

FDDI II The proposed ANSI standard to enhance FDDI. FDDI II will provide isochronous
transmission for connectionless data circuits and connection-oriented voice and video
circuits.

Fiber Distributed Data Interface An emerging high-speed networking standard. The underlying medium is fiber optics,
(FDDI) and the topology is a dual-attached, counter-rotating Token Ring. FDDI networks can
often be spotted by the orange fiber “cable.” The FDDI protocol has also been adapted
to run over traditional copper wires.
Fibre Channel Arbitrated Loop This is one of the possible physical topologies of Fibre Channel. In this topology, the
(FC/AL) Fibre Channel is connected in a loop with devices all connecting to the loop. It can be
thought of as a similar structure to a token ring network. Up to 126 nodes can be
connected to the loop.
Fibre Channel Fabric This is a structure where addressing of ports on a network of Fibre Channel is made
independently of the physical location or address of the target port. Switches are
responsible for passing Fibre Channel packets to the target port regardless of which
Fibre Channel loop or switch the port physically resides on. Multiple switches can be
connected to create large networks with up to 224 addressable ports.
Fibre Channel Point-to-Point This topology for Fibre Channel provides a simple direct connection between just two
nodes. This is the closest approximation to the traditional SCSI topology.
Fibre Channel Topology A number of possible topologies exist for the connection of Fibre Channel. One is point-
to-point, where a direction connection is made between nodes on the Fibre Channel.
Another is Arbitrated Loop where multiple nodes can be connected together. Lastly
there is a Fibre Channel Fabric topology which allows for multiple loops and point-to-
point connections to be combined into networks using switches.

374
The SAN Book III – October 22, 2001

File Allocation Table (FAT) Data written to a magnetic disk is not necessarily placed in contiguous tracks. It is
usually divided into many clusters of data in many locations on the disk surface. The
FAT is the special area on a disk which keeps track of where clusters of data have been
written for later retrieval later.
File and Software Distribution There are many levels to this application set. The simplest is the ability to recognize
revision levels of software installed on the network and to broadcast upgrades so that
revision levels can be maintained automatically.
File Migration File migration is an automated process where candidate files are removed from primary
storage, leaving a phantom (or stub) file in place of each file. Candidate files are
automatically identified for removal based on their inactivity or file type. When the
phantom file is read, it triggers a transparent retrieval from the secondary storage pool
File Retrieval The return of migrated files from the lower level of the storage hierarchy to active
storage in response to user access request.

File Server Local Area Network (LANs) were invented to allow users on the LAN to share and
thereby conserve the cost of peripherals (printers, modems, scanners) and to likewise
share software. The file server is the machine on the LAN where the shared software is
stored.
File System Part of an OS that tracks, retains and monitors all files. The FS can be assessed by other
parts of the OS such as the computer back plain and storage along with applications via
well-documented APIs.
File Tracking The ability of the on-line catalogue to provide complete file versioning and history
information along with specific media locations.

File Transfer Protocol (FTP) (1) An IP application protocol for transferring files between network nodes. (2) An
Internet protocol that allows a user on one host to transfer files to and from another host
over a network.

File Transfer, Access, and The OSI remote file service ad protocol.
Management (FTAM)

File-Oriented Backup Any backup software that instructs the computer to store information in files just as they
appear on the originating computer, making restoration easier and more logical.
FINL_Port An NL_Port which is capable of providing certain Fabric services to other NL_Ports on a
Loop in the absence of a Fabric. This NL_Port will respond to requests to open
communication with AL_PA hex’00’, even though it may actually have another value for
its AL_PA.
FL_Port Fabric-Loop port. An F_Port which is capable of supporting an attached Arbitrated Loop.
An FL_Port on a Loop will have the AL_PA hex’00’, giving the Fabric highest priority
access to the Loop. An FL_Port is the gateway to the Fabric for NL_Ports on a Loop.
Flow Control Flow control is a method of exchanging parameters between two connected devices and
managing the rate of frame transmission. Fibre Channel uses credit-based flow control
to prevent the possibility of frames being transmitted so rapidly that the buffers overflow
and lose data. Flow control. Enables concurrent multiple exchanges. Some Classes of
Service use end-to-end flow control to ensure that lost frames are detected.

375
The SAN Book III – October 22, 2001

Format The arrangement of data on a storage medium.


Frame (1) A frame is the most basic element of a message in Fibre Channel data
communications. It consists of a 24-byte header and zero to 2112 bytes of data. (2) The
basic unit of communication between two N_Ports. Frames are composed of a starting
delimiter (SOF), a header, the payload, the Cyclic Redundancy Check (CRC), and an
ending delimiter (EOF). The SOF and FOF contain the Special Character and are used to
indicate where the frame begins and ends. The 24-byte header contains information
about the frame, including the S_ID D_ID, routing information, the type of data
contained in the payload, and sequence/exchange management information. The
payload contains the actual data to be transmitted, and may be 0-2112 bytes in length.
The CRC is a 4-byte field used for detecting bit errors in the received frame.

Frame Relay High-performance interface for packet-switching networks. Considered more efficient
than X.25 which it is expected to replace. Frame relay technology can handle “bursty”
communications that have rapidly changing bandwidth requirements.

Frame Switch A device, similar to a bridge, which forwards frames based on the frame’s layer 2
address. Frame switches are generally of two basic forms, cut-through switch (on the
fly switching), or store and forward switch. LAN switches such as Ethernet, Token Ring
and FDDI switches are all examples of frame switches.
Full Backup A backup of all files on a storage device. If files or directories have been excluded from
backups, they will not be contained in a full backup.
Gigabit (Gb) One billion bits.
Gigabyte (GB) (1) A billion (slightly more) bytes of data, or a thousand megabytes. Presently, gigabytes
are in the stratosphere of storage capability, but imaging applications commonly take up
huge amounts of data. For example, it only takes ten 8 1/2” by 11” color pictures,
9
scanned at 600 dpi, to fill a gigabyte. (2) Shorthand for 1,000,000,000 (10 ) bytes. Some
9
prefer to use the (10 ) convention commonly found in I/O-related literature rather than
30
the 1,073,741,824 (2 ) convention sometimes used in describing computer system
memory.
Graphical User Interface (GUI) A type of user interface that includes a combination of graphics and the use of pointing
devices, menu bars, overlapping windows and icons.

Groupware A network-based application that lets users collaborate.

GUI Graphic User Interface.

Hardware The mechanical, magnetic and electronic components of a system, e.g., computers,
telephone switches, terminals and the like.
HBA See Host bus adapter.

Hierarchical File System (HFS) In DOS, the file management system that allows directories to have subdirectories, and
sub-subdirectories. In Macintoshes files may be placed into folders, and folders to be
placed within other folders.

376
The SAN Book III – October 22, 2001

Hierarchical Storage Management The process of automatically storing data on the lowest cost devices that can support
(HSM) the performance required by the applications. To users, data storage never fills and file
access, regardless of location in the storage hierarchy, is completely transparent. The
software automatically manages multiple levels of storage hierarchy primary, secondary
(nearline), or off-line
Host Backup Server The backup server that is the central repository (vault) in an enterprise data
management architecture. Local backup servers ship data to the host backup server.
Centralized administration is performed and controlled from the host backup server. The
host backup server can be located anywhere in the enterprise and on any platform in
which will support the requirements.
Host Bus Adapter A host bus adapter allows to connect a server system bus with an external device.
Host Computer Any computer system to which disks are directly attached and accessible. Mainframes,
and servers, as well as workstations and personal computers, can all be considered
host computers in the context of this book, as long as they have disks attached to them
Host-Based Array A disk array whose Array Management Software executes in its host computer(s) rather
than in a disk subsystem.
Hot Backup A backup performed on a local disk drive as opposed to (or in addition to) the classical
backup to tape. It allows faster and more frequent data transmission, and quicker
recovery.
Hot Spare Or On-line spare. A drive that is present in the system but normally unused until another
drive fails, at which time the hot spare is automatically substituted for the failed drive.

Hot Swap The substitution of a replacement unit (RU) in a disk sub-system for a defective one,
where the substitution can be performed while the subsystem is running (performing its
normal function). Hot swaps are manual performed by humans
Hub This is a simple connectivity device that allows for devices to be connected to a fibre
channel loop by being attached to a hub port. The advantage of this is that failures of a
single device on the loop can be isolated from the other ports on the loop. The
aggregate bandwidth of the hub is still that of a single fibre channel loop however.
IDE/Integrated Device (Drive) A common hard drive interface built into most 386/486 PCs. This interface has been
Electronics replaced by EIDE in Pentium and newer PCs.
I/O - Input/output Input/Output. Term applied to a channel or device ( such as an EIDE, parallel port or
SCSI connection) that can perform an input process and an output process at the same
time.
I/O Driver A host computer software component (usually part of the operating system) whose
function is to control the operation of peripheral controllers or adapters attached to the
host computer. I/O drivers communicate between applications and I/O devices, and in
some cases may participate in data transfer, although this is rare with disk drivers,
since most disk adapters and controllers contain hardware for data transfer.

377
The SAN Book III – October 22, 2001

I/O Load The sequence of I/O requests made to one or more disk subsystems by a host
computing environment. The host computing environment includes both applications
and host overhead functions, such as swapping, paging, and file system activity.
I/O Load Balancing The adjustment of system and/or application components so that application I/O
demands are spread evenly across the I/O system’s physical resources. I/O load
balancing may be done manually (by a human) or automatically (by some means that
does not require human intervention).
I/O Operations per Second (IOPS) A generic measure of I/O performance. To be meaningful, the type and operation
mixture of I/O must be specified as well, such as read IOPS.
I/O-Intensive A characterization of applications. An I/O-intensive application is one whose
performance depends strongly on the performance of the resources that provide its I/O
services.
Incremental Backup An operation that backs up all data that has been modified or added since a given date.
The date is usually the date of the last full or incremental backup.

Input/Output (I/O) Refers to the process, techniques and media used for human/machine communication.
Also refers to data submitted between computing components.
Intermix A mode of service defined by Fibre Channel that reserves the full Fibre Channel
bandwidth for a dedicated Class 1 connection but also allows connectionless Class 2
and 3 traffic to share the link if the bandwidth is available.

Internetwork Packet Exchange, LAN protocol developed by Novell for NetWare.


Network Protocol (IPX)

IP Number/IP Address Also referred to as a “dotted quad.” A code number consisting of 4 parts separated by
dots, e.g., 165.113.245.2. Every computer on the Internet has its own unique IP
number.
IPI Intelligent peripheral interface.
Isochronous Transmission Data transmission which supports network-wide timing requirements. A typical
application for isochronous transmission is a broadcast environment which needs
information to be delivered at a predictable time.
JBOD Just a Bunch of Disks. Refer to a disk drive configuration in which there is no
redundancy.

Jukebox A device that holds multiple optical disks and one or more disk drives, and can swap
disks in and out of the drive as needed. Same as an autochanger.
L_Port Loop port. Generic term for an NL_Port or FL_Port i.e., any Fibre Channel port which
supports the Arbitrated Loop topology.
Large I/O Request An I/O, read, or write request that specifies the transfer of a large amount of data.
Large Read Request ‘Large’ usually depends on the context, but often refers to 64 KBytes or more.
Large Write Request

378
The SAN Book III – October 22, 2001

Latency A measurement of the time it takes to send a frame between two locations. Low latency
is a fundamental requirement for storage applications and is typical of I/O channel
technologies. Fibre Channel connections are characterized by low latency. In a disk
drive, latency relates to the disks rotational speed and must be considered in
determining a disk drives total access time.
Library Manager A library manager sits as the piece of middleware between a backup, archive, or some
other application and an automated media repository such as an optical jukebox or tape
library. The library manager intercepts the call for a specific file, looks up the files’
location (which piece of media, surface block ID) and issues commands to the jukebox
or library manager to mount a specific media and retrieve the data requested. The
library manager contains the file history databases, media location database, device
drivers for the supported media repositories and other management functions. Media
management capabilities are usually built into a library manager. It is a superset of
functions to a media manager.
Life Cycle Management A process of tracking and monitoring the number of times that the media is recorded or
certain zones are overwritten (this translates to the number of passes over the heads).
Management also requires the monitoring of raw error rates in order to prevent
excessive usage and data loss.

Local Area Network (LAN) (1) A network covering a relatively small geographic area (usually not larger than a floor
or small building). Compared to WANs, LANs are usually characterized by relatively high
data rates. (2) Network permitting transmission and communication between hardware
devices, usually in one building or complex. (3) High-speed transmissions over twisted
pair, coax, or fiber optic cables that connect terminals, personal computers, mainframe
computers, and peripherals together at distances of about one mile or less.
Local Backup Server A CPU and storage pool which acts as a centralized backup device and repository for a
set of workstations and servers in a network. Multiple local backup servers are tied
together and act synchronously to provide backup for the network under central control.
Data is ported offsite to the host backup server. Administration is performed, centrally,
for all local backup servers either locally or from the host backup server.
Logical Disk A set of contiguously addressed member disk blocks that is part of a single virtual disk-
to-member disk mapping. Logical disks are used in some array implementations as
constituents of logical volumes or partitions
Logical Volume A virtual disk made up of logical disks. Also called a virtual disk, volume set, or partition.

Magneto-Optic (MO) Recording Recording data using optical means to change the polarity of a magnetic field in the
recording medium. Data is erasable and/or rewritable.
Managed Hub This is a technique for providing statistics information about the traffic on a hub. No
actual management of the hub is usually possible using this interface, but information
and notification of failures can be achieved. This interface often uses Simple Network
Management Protocol (SNMP) Management Information Bases (MIBs) as a standard
protocol for providing this information.

379
The SAN Book III – October 22, 2001

Mapping The conversion between multiple data addressing schemes. Most commonly used to
refer to conversion between member disk block addresses and block addresses of the
virtual disks presented to the operating environment by Array Management Software.

Mean Time Between Failures A measure of equipment reliability, the higher the MTBF, the more reliable the
(MTBF) equipment.

Media Plural of medium. The physical environment through which transmission signals pass.
Common network media include twisted pair, coaxial and fiber optic cable, and the
atmosphere (through which microwave, LASER, and infrared transmission occurs).
Media Management The software will monitor the number of write passes or use other techniques to
monitor media quality, aging, and recommend new media and/or suggest using the
media for another purpose such as archiving. It is a pro-active and automated feature
allowing preventive media failures and requiring less system administrator time in
manual media management and error control.
Media Types Managed Optical or tape media are common. There are many types of tape media that are
supported by libraries; 3490, DLT. 8 MM. 4 MM, and VHS (VLDS).

Megabits Per Second (Mbps) A digital transmission speed of millions of bits per second.

Megabyte A million bytes. A thousand kilobytes.


6 6
Megabyte, Mbyte (MB) Shorthand for 1,000,000 (10 ) bytes. Pronounced megabyte. This book uses the 10
20
convention commonly found in I/O related literature rather than the 1,048,576 (22 )
convention commonly used to describe computer system memory.
Metadata Data that describes data. In disk arrays, metadata often consists of items such as array
membership, member segment sizes, descriptions of logical disks and partitions, and
array state information.
Mirroring A form of RAID in which Array Management Software maintains two or more identical
copies of data on separate disks. Also known as RAID Level 1 and disk shadowing.
Mirrors Common terms for a disk array that implements RAID Level 1, or mirroring technology.
Mirrored Disks
Mirrored Array

MTBF Mean Time Between Failure; a measure of reliability.


Multicast Sending a copy of the same transmission from a single source device to multiple
destination devices on a fabric.
Multiple Parallel Backup The ability to manage multiple jobs (backup/restore) at the same time by keeping
multiple tape drives operating in parallel.
Multiple Virtual Storage (MVS) One of the family of IBM operating systems for the System/370 or System/390
processor, such as MVS/ESA. MVS is one of the supported server environments.
Multiplex The ability to intersperse data from multiple sources and destinations onto a single
transmission medium.
N_Port Node port. A port on a computer, disk drive, etc. through which the device does its Fibre
Channel communication.

380
The SAN Book III – October 22, 2001

Near-Line Data that is available on a secondary storage device that the user can access, but at a
slower rate than the on-line data is accessed.
Nearline Storage Or Near Online Storage. Storage that can be accessed online only through the use of
robotics. It does not require human intervention, and can be accessed much quicker
than off-line storage, though not as quickly as online prime storage.

Network (1) A collection of computers and other devices that are able to communicate with each
other over some network medium. (2) An aggregation of interconnected nodes,
workstations, file servers, and/or peripherals, with its own protocol that supports
interaction.

Network Analyser A hardware/software device offering various network troubleshooting features, including
protocol-specific packet decodes, specific pre-programmed troubleshooting tests,
packet filtering, and packet transmission.
Network Attached Storage (NAS) This is the provision of storage in a form that is readily accessible on a network.
Traditionally this meant having a storage array device that incorporated a file system
and presented either a networked file system protocol such as NFS, or else emulated a
disk device so that the array could be connected to a storage I/O interconnect to the
host.
Network Availability A measure of network uptime. High network availability is a term designed to convey
the message that 100% network uptime is very desirable. Most networks only achieve
96% uptime due to planned and unplanned outages.
Network Backup Software Backup software capable of providing backup and restore of network files for servers
and clients. This software is capable of backup and restore only. It is flexible and has
many features such as tape catalogue s, tape libraries, and many storage devices.
Network Data Management Backup software which includes high performance backup and restore features plus
Software manages data and files in the network. Features such as disk-based database/librarian,
automated archiving, and file migration set it apart from Network Backup.
Network File System (NFS) (1) A protocol defined by Sun Microsystems that that extends TCP/IP
network file services. NFS permits remote node files to appear as though
they are stored on a local workstation. (2) A distributed file system in
UNIX developed by Sun Microsystems which allows a set of computers
to cooperatively access each other’s files in a transparent manner.
Network Storage Management (1) Network Storage Management is not an application. Rather, it is a family of related
processes and application that manage storage devices as network resources. Storage
management applications range from asset management to hierarchical storage
management. (2) A system responsible for managing at least part of a network. NMSs
communicate with agents to help keep track of network statistics and resources.
Network Storage Resource A storage pool that can be accessed by every client in the network. Its access is
completely transparent to the user, whether the storage is online or nearline.

Network Topology Physical arrangement of nodes and interconnecting communications links in networks
based on application requirements and geographical distribution of users.
NL_Port Node-Loop port. An N_Port which can operate on the Arbitrated Loop topology.

381
The SAN Book III – October 22, 2001

NOS Network Operating System; UNIX, WIndows NT, and Novell Netware are examples.

Off-Line Data that is not physically stored on an accessible drive, such as removable tapes or
disks.

On-Line Data that is available on a primary storage device so that it is readily accessible to the
user.
Open File Backup The backup of open files (which may be modified during backup). The file must remain
static for backup but available for reading and writing.

Open Network Architecture (ONA) The structure of a communications network that permits all types of terminals and users
to have access to any network capability without compatibility restrictions.

Operating system Collection of programs that, taken together, manage the hardware and software; it is
the operating system that makes the hardware usable, providing the mechanisms that
application programs use to interact with the computer.

Optical (1) Containing lenses mirrors, etc., as in optical view-finder and optical printer. (2) In
general, having to do with light and its behavior and control, as in optical properties,
optical rotation. (3) Pertaining to the science of light and vision.

Optical Disk A storage device that is written and read by laser light. Certain optical disks are
considered Write Once Read Many (WORM), because data is permanently engraved in
the disk’s surface either by gouging pits (ablation); or by causing the non-image area to
bubble, reflecting light away from the reading head. Erasable optical drives use
technologies such as the magneto-optic technique, which electrically alters the bias of
grains of material after they have been heated by a laser. Compact disks (CDs) and
laser (or video) disks are optical disks. (When referring to CD technology, the spelling
disc is used. In all other cases, the spelling disk is used.)

Optical Drive Machine for reading or writing a data storage medium (disk, tape, card, etc.) that uses
light for examining patterns.
Parallel Access Array A disk array in which the data mapping and protection model assumes that all member
disks operate in unison, and that all member disks participate in the execution of every
application 1/0 request. Parallel access arrays may require physical disk
synchronization, or they may approximate true parallel access behavior.
Parallel/Concurrent Parallel is the capability to manage multiple jobs (backup/restore) at once by keeping
multiple tape drives operating in parallel. Concurrence allows a storage device to
receive data from more than one system at a time, interleaving the data streams.
Partial File Backup A backup in which only the changes to a file (usually a large file) are copied to backup
media instead of copying the entire file. This assumes that an older version of the entire
file has already been copied to backup media.
Partition A virtual disk. The term partition is most often used when a redundancy group is
presented to a host operating environment as more than one virtual disk or volume set.
Also used in complex arrays with dynamic mapping to denote a collection of redundancy
groups dedicated to providing the storage for a subset of an array’s virtual disks.

382
The SAN Book III – October 22, 2001

Physical Media Any physical means for transferring signals between OSI systems. Considered outside
the OSI Model, and sometimes referred to as “Layer O,” or the bottom of the OSI
Reference Model.
Point-to-Point Pertaining to a direct link that supports data transmission between two nodes.
Ports Fibre Channel ports come in a number of flavours depending on the topology of the
Fibre Channel

N_Ports are simple equipment node ports in a point-to-point connection topology

NL_Ports are node ports connected to an Arbitrated loop.

F_Ports are point-to-point ports connected to a fabric. Generally this means that
the F_Port is a port on a switch.

FL_Ports are ports connecting from one loop to a switch and thus to a fabric

E_Ports are expansion ports used to interconnect switches together.

G_Ports are classified by some switch companies as ports that can be either
E_Ports or F_Ports depending on usage.
Protocol Set of data communication conventions, including timing, control, formatting, and data
representation. See Transport protocol
QIC/Quarter Inch Cartridge A magnetic tape technology used for backup. QIC tape comes in 3.5-inch minicartridges
and 5.25-inch data cartridges.
RAID Level A number designating the general configuration of an array. RAID configurations are
defined and generally accepted from level 0 through 5.
Random Access The ability to position data (as in disk) without reading intervening data blocks (as in
tape).
Random I/O An I/O load whose consecutively issued read or write requests do not specify adjacently
Random Reads located data. Random I/O is characteristic of I/O request-intensive applications.
Random Writes
Read Cache A cache segment dedicated to saving information read from the disk drives, on the
assumption that this data will be soon requested again by the system. The system will
initiate further disk accesses only if the desired information is not located in the cache.

Read Only Memory (ROM) Data stored in a medium that allows it to be accessed but not erased or altered.
Read/Write Ratio The ratio of read operations to write operations in a typical host system workload. This
is important in selecting RAID configurations, as some configurations are inefficient in
write intensive environments.
Read-Only Access to Migrated Data The normal retrieval process automatically writes the file back to primary storage. If the
user is scanning a large number of files, this procedure can rapidly overload the primary
file system. An alternative is to read the files from the migration storage pool directly
into memory and not write to disk unless the user chooses to do so. This way, the
unused files can be discarded without impacting disk storage.

383
The SAN Book III – October 22, 2001

Reconstruction The regeneration of all of the data from a failed disk in a RAID Level 1, 3, 4, 5, or 6 array
onto a replacement disk. Reconstruction normally occurs without interruption of
application access to data stored on the array’s virtual disks.
Reduced Mode A mode of RAID array operation in which not all components of the array are
functioning, but the array as a whole is functioning properly (responding to application
read and write requests).
Reduction The removal of a member disk from a RAID array. Reduction most often occurs because
of member disk failure, however, some RAID implementations allow reduction for
system management purposes.
Redundancy The utilization of modules that are not necessary for a normal system operation. They
get substituted to (or can perform the same function as) another component, allowing
the system to remain operational if the component fails.

Redundant Arrays of Inexpensive or A storage device that uses several optical discs working in tandem to increase
Independent Discs (RAID) bandwidth output and to provide redundant backup.
Regeneration The process of rebuilding user data that was stored on a failed member disk of a RAID
Level 1, 3, 4, 5, or 6 array . Regeneration may be used to recover data when a member
disk has failed. It may also be used to recover data when an unrecoverable media error
is encountered on a member disk. Data regeneration in a RAID Level I array consists of
delivering the alternate copy of the data. Data regeneration in a RAID Level 3, 4, or 5
array consists of executing the array’s parity computation algorithm. Data regeneration
in a RAID Level 6 array consists of choosing the more convenient of the two parity
algorithms, rebuilding the data, and adjusting the alternate parity algorithm as required.

Relational Database A database built and operated in accordance with the relational model of data which
holds that all data be organized as a set of two dimensional arrays or tables which have
a relation to each other.

Remote Access The process of allowing remote workers to access a corporate LAN over analog or
digital telephone lines.

Remote Access Server Access equipment at a central site that connects remote users with corporate LAN
resources.

Remote File System (RFS) A distributed file system, similar to NFS, developed by AT&T and distributed with their
UNIX System V operating system.
Remote Vaulting The software has the ability to reach out or transmit over phone or direct connection
with other sites, and manage that data, essentially treating remote site data as a client
system or visa versa.
Remote Vaulting Capability The ability of the media manager to transmit files and data offsite to a remote vault.
Replacement Disk A disk available for use as or used to replace a failed member disk in a RAID array.

384
The SAN Book III – October 22, 2001

Replacement Unit A component or collection of components in a disk subsystem which are always
replaced (swapped—q.v.) as a unit when any part of the collection fails. Abbreviated
RU. A replaceable unit is commonly called a consumer replaceable unit (CRU) if the
replacement can be performed by the end user, and a field replaceable unit (FRU) if the
replacement must be performed by a vendor service representative. Replacement units
may be cold, warm, hot, or auto-swapped (q-v-).
Reporting The ability to develop reports on access, media status, and other management requests.
Request-Intensive A characterization of applications. Also known as I/O -intensive
Restore A function that allows users to copy files from the backup storage pool to an online
storage device.
Retrieve A function that allows users to copy files from the archive storage media to an online
storage device.

Rewritable Optical Optical media from which data can be erased and new data added. Magneto-optical
and phase change are the two main types of rewritable optical disks.

Router (1) An OSI Layer 3 device that can decide which of several paths network traffic will
follow based on some optimality metric. Also called a gateway (although this definition
of gateway is becoming increasingly outdated), routers forward packets from one
network to another based on network-layer information. (2) A dedicated computer
hardware and/or software package which manages the connection between two or
more networks.

Routing Bridge MAC-layer bridge that uses network layer methods to determine a network’s topology.
Scalable Capable of growing through additions of modular increments, without necessitating
major modifications to the original system. In a storage context, a scalable system is
capable of efficiently handling the requirements from a small system to a large system,
in terms of capacity, performance, availability, connectivity, and/or manageability.
SCSI Small Computer Systems Interconnect. A collection of ANSI standards which define I/O
buses primarily intended for connecting mass storage subsystems or devices to hosts
through I/O bus adapters. Originally intended primarily for use with small (desktop and
deskside workstation) computers, SCSI has been extended to serve most computing
needs, and is arguably the most widely implemented I/O interconnect in use today. SCSI
is used for local data communication between a host CPU and an attached SCSI bus that
contains intelligent peripheral devices such as disks, tapes, scanners and printers. As
many as six devices can be connected to a single SCSI channel. Variations include:

Narrow SCSI, which transfers 8 bits in parallel;

Wide SCSI, a higher-performing SCSI interface that transfers 16 bits in parallel;

SCSI, which has a transfer rate of 5MB/sec;

SCSI-2, a higher performance SCSI interface with transfer rates up to 10MB/sec on


an 8-bit interface and 20MB/sec on a 16-bit interface; and

Ultra SCSI, a faster SCSI-2 interface that transfers 20MB/sec for narrow 8-bit
versions and 40MB/sec for wider 16-bit versions.

385
The SAN Book III – October 22, 2001

Security Features Security features include operator access privileges, authorization levels, and password
control to limit access to unauthorized files and data.
Seek The physical motion to relocate the arm from one track to another.
Selective Backup A function that allows the backup program to include or exclude certain files from being
systematically backed up during a full or an incremental backup operation.
Sequence A series of related frames, when strung together in numbered order, creates a sequence
which can be transmitted over a Fibre Channel connection as a single operation
Sequential I/O An 110 load consisting of consecutively issued read or write requests to adjacently
Sequential Reads located data. Sequential I/O is characteristic of data transfer intensive applications.
Sequential Writes
Server A computer which is dedicated to one task. A database or directory server would be
responsible for responding to a user’s search request, returning the list of stored
documents that meets with the parameters of the request.
Shelf Storage Media placed external to an automated repository but, nearby so that an operator can
mount it upon request.

Simple Network Management (1) The Internet network management protocol. SNMP provides a means to monitor and
Protocol (SNMP) set network configuration and runtime parameters. (2) A standard protocol that runs
over an IP link to provide management of network type devices without performing
continual polling.

Small Computer System Interface An industry standard for connecting peripheral devices and their controllers to a
(SCSI) microprocessor. SCSI defines both hardware and software standards for communication
between a host computer and a peripheral. Computers and peripheral devices designed
to meet SCSI specifications should work together. A single SCSI adapter card plugged
into an internal IBM PS/2 micro channel PC slot can control as many as seven different
hard disks, optical disks, tape drives and scanners, without siphoning power away from
the computer’s main processor. Formerly known as SASI (Shugart Associates Systems
Interface).
Small I/O Request An I/O, read, or write request that specifies the transfer of a relatively small amount of
Small Read Request data. ‘Small’ usually depends on the context, but most often refers to 8 KBytes or fewer.
Small Write Request
Solid State Disk A disk whose storage capability is provided by solid state random access memory rather
than magnetic or optical media. An electronic storage element generally offers very high
performance, because seek and rotational latency times are negligible, and may offer
very high data transfer rate compared to a conventional disk. Cost per byte of storage,
however, is quite high compared to conventional disks, and volumetric density is lower.
These two factors combine to make solid state disk generally lower capacity than
magnetic or optical disks.
Spare, Spare Disk Informally, a disk or P_extent reserved for the purpose of substituting for a like entity in
case of failure of that entity.

SSA (Serial Storage Architecture) A high speed serial interface designed and marketed by IBM.

386
The SAN Book III – October 22, 2001

Storage Area Network (SAN) Simply put, this is the evolutionary step of connectivity provided by a support (e.g., Fibre
Channel) between host and storage. This connectivity provides a true network of storage
devices and the hosts that access the storage.

Storage Capacity Amount of data that can be contained in an information holding device or main memory,
generally expressed in terms of bytes, characters or words.
Storage Density Usually refers to recording density (bpi, tpi, or a combination of both).
Storage Hierarchy A logical ordering of the backup and archive storage devices, as defined by an
administrator with system privilege. Generally, the ordering is based on the speed and
capacity of the devices. Typically, each storage pool is associated with a different type
of storage device; online-disk, nearline-optical, jukebox tape library or offline-tape.

Storage Media The physical device itself, onto which data is recorded. Magnetic tape, optical disks,
floppy disks are all storage media.
Storage Pool A named collection of storage volumes whether online, nearline, or off line.
Storage Subsystem A disk subsystem (q.v.) whose intelligent controller or I/O bus adapter is capable of
attaching other forms of mass storage such as CDROMs or tape transports.

Striped Array The collection of disks comprising an array which implements the RAID Level 0 or disk
striping (q.v.) mapping technique.
Striping Short for disk striping (q.v.); also known as RAID Level 0. The process of recording data
across multiple disks rather than on one drive. Data is divided into segments each of
which is written to successive drives. Striping is also referred to as RAID 0.

Stub File The record of the original file left on the primary drive after the file itself has been
migrated to a secondary drive. Also called a key or a place-holder. Click on it and the
file appears on your screen as it normally would, but delivered to you at the slower
access speed of the secondary device. The stub file takes up much less space than the
original, unmigrated file. When the user attempts to open a stub file, it copies back the
migrated material to the primary drive, then opens it.
Subsystem-Based Array A disk array whose Array Management Software executes in the controller or host I/O
bus adapter of an intelligent disk subsystem.
Swap The installation of a replacement unit in place of a defective unit. Units are parts of a
disk subsystem which may either field replaceable (FRUs) by a vendor service
representative or consumer replaceable (CRUs). The swap operation itself may either be
a cold, warm, hot, or auto-swap, depending on the state in which the disk subsystem
must be in order to perform it.
Switch — Switched Fabric A Fibre Channel switch routes the data . It has the ability to shift circuit connections as
required to support data transmission requests. When this type of connection is
expanded to include more than one switch and multiple nodes, it is called a switched
fabric

Switched Ethernet Configuration supporting an Ethernet hub with integrated MAC layer bridging or
switching capability to provide each port with 10 or 100 Mbps of bandwidth. Separate
transmissions can occur simultaneously on each port of the switching hub, and the
switch filters traffic based on the destination MAC address.

387
The SAN Book III – October 22, 2001

System Disk The disk on which a system’s operating software is stored. The system disk is usually
the disk from which the operating system is bootstrapped (initially loaded into memory).
The system disk frequently contain the system’s swap and/or page files. It may also
contain libraries of common software shared among several applications.

Systems Network Architecture IBM’s very successful means of networking remotely located computers. It is a tree-
(SNA) structured architecture, with a mainframe host computer acting as the network control
center. Unlike the telephone network, which establish a physical path for each
conversation, SNA establishes a logical path between network nodes, and it routes each
message with addressing information contained in the protocol.

Tape Backup Making mag tape copies of hard disk and optical disc files, for disaster recovery.

Tape Drive The machine, actually a collection of devices, that transports, reads and or writes a
magnetic tape.
TCP/IP Transmission Control Protocol/Internet Protocol. A set of communication protocols that
support peer-to-peer connectivity functions for both local and wide area networks. Also
used for Internet access from the desktop.

Terabyte (TB) From “tera,” which means trillion, although it actually means 1,099,511,627,776 bytes
in a computer’s binary system. A terabyte is 1,024 gigabytes.
Throughput (1) The amount of time it takes for the processing of data from the beginning of a
particular process, to the end of the process. Throughput also can refer to the number
of items completed in the process. (2) The number of I/O requests satisfied per unit
time. Expressed in requests/second, where a request is an application request to a
storage subsystem to perform a read or write operation.
Throughput-Intensive A characterization of applications. Also known as request-intensive. A throughput-
intensive application is an I/O intensive (q.v.) application which makes large numbers of
requests asynchronously, as opposed to a data transfer-intensive (q.v.) application.

Transmission Control (1) A set of protocols developed by the Department of Defense to link dissimilar
Protocol/Internet Program (TCP/IP) computers across networks. (2) A reliable, full duplex, connection-oriented end to end
transport protocol running on top of IP.

Transparent Retrieval Transparent retrieval is achieved by several processes. The objective is for the user to
access all their files and data as if they were on-line. Some methods leave a stub or
phantom file which have the same name as the migrated file. When the application
begins to read the file, the stub triggers a retrieval. Control is returned to the user along
with a message so that they can do something else while the retrieval occurs.
Travan A commonly used tape backup technology that grew out of the QIC standard. Because
of their simplicity and affordability, Travan products tend to be ideal for backup desktop
and portable PCs and smaller servers.

388
The SAN Book III – October 22, 2001

Transport Protocols A set of rules and procedures that support connectivity functions for local or wide area
networks. Examples are:

TCP/IP (UNIX)

IPX/SPX (NetWare)

SNA (IBM)
UDF/Universal Disk Format A file system for CD optical media. UDF was designed for read-write interoperability
between all the major operating systems as well as compatibility between CD rewritable
and CD write-once media.
Unformatted Capacity Storage capacity of disk drive prior to formatting; also called the gross capacity.

UNIX A general-purpose, multi-user, multitasking operating system invented by AT&T. UNIX is


powerful and complex, and needs a computer with a large amount of RAM memory to
support its power. UNIX allows a computer to handle multiple users and multiple
programs simultaneously. And it works on many different computers, which means you
can often take applications software and move it, with little changing, to a bigger,
different computer, or to a smaller, different computer. This process of moving
programs to other computers is known as “porting.” UNIX uses TCP/IP as its standard
communications protocol.

Value Added Reseller / Companies that buy equipment from computer or telephone manufacturers, add some
Value Added Dealer of their own software and possibly some peripheral hardware to it, and then resell the
(VAR/VAD) whole computer or telephone system to end users.
Virtual Disk An alternative name for volume set (q.v.). In order to be consistent with ANSI X3T9.2
RAID Study Group terminology, the RAID Advisory Board is gradually replacing this
commonly used term with the term volume set in its publications.

Volume Label A name assigned to a floppy or hard disk in MS-DOS. The name can be up to 11
characters in length. You assign a label when you format a disk or, at a later time, using
the LABEL command.
Volume Management The ability to make two distinct disk drives or tape subsystems appear to the OS as one
logical system. Therefore, if you have 4 disks drives, you can structure those four disks
drives as one large disk farm enabling data to be written to the entire disk farm at the
same time.
Volume Set (An ANSI X3.T10 term) A contiguous range of logical block addresses presented to the
operating environment by Array Management Software. The volume set, sometimes
called a virtual disk (q.v.), is the array entity which most closely resembles a disk when
viewed by the operating environment.

Web Browser Client software that requests and displays HTML documents and other Internet or
intranet resources.

Wide Area Network (1) A network which encompasses interconnectivity between devices over a wide
(WAN) geographic area. Such networks would require public rights-of-way and operate over
long distances. (2) A network linking computers, terminals and other equipment that
covers an area larger than a single building or campus.

389
The SAN Book III – October 22, 2001

Workstation A desktop computer that performs local processing and accesses LAN or WAN services.
Write Cache A cache segment used to accumulate data before writing to the drive on the theory that
a single large write operation is more efficient than several smaller transfers.
Write Once Read Many Optical storage device on which data is permanently recorded. Data can be erased, but
(WORM) not altered, and no additional data can be added.
Write Penalty The low apparent performance of application write requests to independent access RAID
arrays’ virtual disks. The write penalty is inherent to independent access RAID data
protection schemes, which require multiple member writes for each application write
and ranges from minimal (RAID Level 1) to substantial (RAID Levels 5 and 6). Many array
designs include features such as write-back cache specifically to minimize the write
penalty.
Zoning This is the term used by some switch companies to denote the division of a SAN into
sub nets that provide different levels of connectivity or addressability between specific
hosts and devices on the network. In effect routing tables are used to control access of
hosts to devices. This zoning can be performed by cooperative consent of the hosts or
can be enforced at the switch level. In the former case, hosts are responsible for
communicating with the switch to determine if they have the right to access a device.

390
The SAN Book III – October 22, 2001

Table of Figures

Figure 1. PC Unit Shipments (1981–99) ...................................................................................................7


Figure 2. PC Shipments to Large Customers ...........................................................................................8
Figure 3. Storage Industry Projections ...................................................................................................10
Figure 4. Total Worldwide Server and RAID-based Storage Revenues ..................................................10
Figure 5. Data Storage and the Economic Pullback................................................................................11
Figure 6. Digital Camera Usage .............................................................................................................14
Figure 7. Online Photo Sites ..................................................................................................................16
Figure 8. Worldwide Digital Cameras .....................................................................................................17
Figure 9. Our Featured Storage End Users ............................................................................................18
Figure 10. Intuit’s SAN ROI ....................................................................................................................20
Figure 11. Annual Storage Demand for GSFC EOS Satellite System.....................................................23
Figure 12. Lawrence Livermore National Laboratories’ Projected Disk and Tape Storage Demand........24
Figure 13. Barry Bonds of the San Francisco Giants: An Unlikely Storage End User..............................27
Figure 14. Potential Storage Demand for Major League Baseball ..........................................................28
Figure 15. Pacific Bell Park: E-turnstiles, 40,000 seats, and 7 Terabytes Storage Capacity ...................29
Figure 16. Example of the Demand Effect of Increased Capacity Utilization ...........................................31
Figure 17. Demand Effect of Increased Storage Capacity Utilization ......................................................32
Figure 18. Data Replication Software Revenues ....................................................................................33
Figure 19. Price Trends of External Hardware RAID Storage .................................................................34
Figure 20. Worldwide Forecast for RAID Storage (in Terabytes Shipped) ..............................................37
Figure 21. Revenue Growth of ERP and Business Applications .............................................................38
Figure 22. Worldwide Data Warehouse Software Revenue Growth Projections .....................................39
Figure 23. E-mail Messages Sent on an Average Day ...........................................................................40
Figure 24. Emerging Application Growth Projections..............................................................................40
Figure 25. Emerging Application Growth Projections..............................................................................41
Figure 26. North American Enterprise Video Streaming Market Opportunity ..........................................42
Figure 27. Corporate Bandwidth Penetration..........................................................................................43
Figure 28. Storage Opportunity in Streaming Video................................................................................44
Figure 29. Previous “Killer Application” Adoption Rates..........................................................................44
Figure 30. Size Terminology: From Byte to Yottabyte............................................................................45
Figure 31. Sizing for Storage Applications..............................................................................................45
Figure 32. High Availability Metrics ........................................................................................................46
Figure 33. Downtime Costs ....................................................................................................................46
Figure 34. Downtime Costs — Companies.............................................................................................46
Figure 35. Traditional DAS (Direct Attached Storage) Architecture.........................................................48
Figure 36. SAN Architecture...................................................................................................................49
Figure 37. SAN Market Projections ........................................................................................................50
Figure 38. RAID-Based Disk Storage in SAN, NAS, and DAS ................................................................51
Figure 39. Comparison of DAS versus SAN ...........................................................................................51
Figure 40. Tape Backup Using DAS.......................................................................................................52
Figure 41. Tape Backup in a SAN ..........................................................................................................53
Figure 42. Better Capacity Utilization in a SAN.......................................................................................53
Figure 43. Many-to-Few “Fan-in” Saves Money or Utilizes Higher-end Subsystems...............................54
Figure 44. Cost Advantage of Centralized Storage.................................................................................55
Figure 45. SAN Is More Scalable ...........................................................................................................56
Figure 46. Any-to-Any Connectivity ........................................................................................................56
Figure 47. Traditional DAS (Direct Attached Storage) Backup................................................................57
Figure 48. Traditional DAS (Direct Attached Storage) Backup Stops when a Server Fails......................57
Figure 49. SANs Create Multiple Paths to Circumvent Server Failures...................................................58
Figure 50. SANs Create Multiple Paths to Circumvent Storage Networking Failures ..............................58
Figure 51. SANs Create Multiple Paths to Circumvent Storage Subsystem Failures ..............................59
Figure 52. Potential SAN Manager .........................................................................................................59
Figure 53. RAID-Based Disk Storage by Architecture.............................................................................61
Figure 54. Traditional NAS Architecture .................................................................................................62
Figure 55. NAS Versus SAN ..................................................................................................................63
Figure 56. Market Positioning of NAS Vendors.......................................................................................65
Figure 57. NAS Revenues and Projections ............................................................................................66
Figure 58. NAS Units and Projections ....................................................................................................66
Figure 59. Total NAS Market Share by Revenue, 2000 ..........................................................................66
Figure 60. Traditional NAS Architecture .................................................................................................68
Figure 61. By Scaling NAS, It Becomes a SAN ......................................................................................68
Figure 62. Using a SAN Behind NAS to Provide Failover and Disk Mirroring..........................................69

391
The SAN Book III – October 22, 2001

Figure 63. Traditional Celerra-Based NAS Architecture..........................................................................70


Figure 64. Potential Celerra Based SAN Architecture.............................................................................71
Figure 65. Data-Centric Architecture ......................................................................................................71
Figure 66. Servers Are Clustered to Provide Greater Reliability .............................................................72
Figure 67. Server Clustering Using Storage Network Switches ..............................................................73
Figure 68. Scalable Storage Networking Cluster ....................................................................................73
Figure 69. Designing a Potential Systems Area Network (san)...............................................................74
Figure 70. Veritas Clustering Revenues .................................................................................................75
Figure 71. SAN Storage Market in 2000 .................................................................................................78
Figure 72. RAID-Based Disk Storage Revenues and Projections ...........................................................79
Figure 73. Total RAID Storage Market Share by Revenue .....................................................................79
Figure 74. EMC’s Stock Price Versus the S&P.......................................................................................80
Figure 75. Major Hard Disk Drive Components ......................................................................................82
Figure 76. Disk Drives............................................................................................................................83
Figure 77. EMC Symmetrix Disk Drive Subsystem .................................................................................84
Figure 78. Different Server/RAID Configurations ....................................................................................85
Figure 79. Internal Versus External Storage ...........................................................................................85
Figure 80. Internal RAID Storage Market Share by Revenue..................................................................86
Figure 81. EMC’s Symmetrix Revenues .................................................................................................87
Figure 82. Internal Storage Versus External Storage Market Share by Revenues (in 2000) ...................87
Figure 83. JBOD Storage Versus External Storage by Revenues in 2000 ..............................................88
Figure 84. Total RAID-Based Disk Storage ............................................................................................88
Figure 85. External RAID Storage Market Share by Revenues (1998 and 2000)....................................89
Figure 86. Mirroring in RAID 1................................................................................................................91
Figure 87. Redundancy in RAID 1 ..........................................................................................................91
Figure 88. Striping..................................................................................................................................92
Figure 89. Parity in RAID 5.....................................................................................................................92
Figure 90. Redundancy in RAID 5 ..........................................................................................................93
Figure 91. RAID Vendor Comparison .....................................................................................................94
Figure 92. Subsystem Architecture.........................................................................................................95
Figure 93. Subsystems Architecture Model ............................................................................................96
Figure 94. Cache Hit Versus Cache Miss ...............................................................................................98
Figure 95. Algorithms Guess Which Files Will Be Accessed Soon..........................................................99
Figure 96. Effective Versus Ineffective Caching .....................................................................................99
Figure 97. Mirrored Cache Versus Single Cache in Cache Failure .......................................................100
Figure 98. Bus Architecture..................................................................................................................101
Figure 99. Switched Versus Bus ..........................................................................................................102
Figure 100. Switched Architecture and Mirrored Cache........................................................................103
Figure 101. Several NAS Configurations Are Possible .........................................................................104
Figure 102. Network Appliance’s F840.................................................................................................105
Figure 103. EMC’s Celerra...................................................................................................................106
Figure 104. Traditional Celerra Architecture .........................................................................................107
Figure 105. EMC’s CLARiiON FC4700.................................................................................................108
Figure 106. Quantum’s Snap 4100.......................................................................................................109
Figure 107. Maxtor’s MaxAttach 4100 ..................................................................................................109
Figure 108. A General Purpose Server Is More Complex — and Expensive — Than a NAS Appliance111
Figure 109. Adding More Storage: Quantum’s Entry NAS Appliances Versus an NT Server or Hard
Drive Addition....................................................................................................................................111
Figure 110. File System and Volume Management Architecture ..........................................................112
Figure 111. Block-Based Storage.........................................................................................................114
Figure 112. The Hierarchy of Data Storage — Volume Management ...................................................115
Figure 113. Volume Management Provides Greater Manageability Than Simple Block Storage...........115
Figure 114. File-Based Storage............................................................................................................117
Figure 115. The Hierarchy of Data Storage — File Systems ................................................................118
Figure 116. File System DAS Architecture ...........................................................................................121
Figure 117. File System SAN Architecture ...........................................................................................121
Figure 118. File System NAS Architecture ...........................................................................................122
Figure 119. Various Data Architectures ................................................................................................123
Figure 120. Storage Subsystem Connectivity by Operating and File System........................................124
Figure 121. Heterogeneous Storage ....................................................................................................125
Figure 122. Network Appliance Uniquely Stores Data in a Heterogeneous, Neutral Format .................125
Figure 123. Network Appliance’s Heterogeneous Storage....................................................................126
Figure 124. Externalizing the File System using EMC’s Celerra ...........................................................127
Figure 125. EMC’s Celerra Without HighRoad .....................................................................................128
Figure 126. EMC’s HighRoad Software Enables Greater Symmetrix Connectivity................................128
Figure 127. EMC’s HighRoads Can Leverage SANs for Best-in-Class Performance ............................129

392
The SAN Book III – October 22, 2001

Figure 128. Virtualization in a Server-Centric Architecture....................................................................130


Figure 129. Virtualization in a Network-Centric Architecture .................................................................130
Figure 130. Virtual Storage Pool of Unlimited Storage on Demand.......................................................131
Figure 131. Core Storage Management and Virtualization Software, UNIX platform.............................136
Figure 132. Financial Power Comparison.............................................................................................137
Figure 133. BMC Software Acquisitions ...............................................................................................137
Figure 134. Computer Associates Acquisitions.....................................................................................137
Figure 135. EMC Corporation Acquisitions...........................................................................................138
Figure 136. IBM Acquisitions................................................................................................................138
Figure 137. Legato Systems Acquisitions.............................................................................................139
Figure 138. Network Appliance Acquisitions.........................................................................................139
Figure 139. Veritas Acquisitions ...........................................................................................................140
Figure 140. Storage Software Trends...................................................................................................141
Figure 141. Total Worldwide Storage Software Market Revenues by Platform .....................................142
Figure 142. Mainframe Storage Software Market Share (2000)............................................................142
Figure 143. UNIX Storage Software Market Share (2000) ....................................................................143
Figure 144. NT Storage Software Market Share (2000)........................................................................143
Figure 145. Total Worldwide Storage Software Market Revenues ........................................................144
Figure 146. Total Worldwide Storage Software Market Revenues by Selected Vendor ........................145
Figure 147. EMC’s ControlCenter Software..........................................................................................148
Figure 148. Prisa Networks’ VisualSAN Network Manager...................................................................150
Figure 149. Prisa Networks’ VisualSAN Performance Manager............................................................151
Figure 150. Storage Resource Management Projections......................................................................151
Figure 151. Storage Resource Management Market Share, All OS ......................................................152
Figure 152. Storage Resource Management Market Share, UNIX........................................................152
Figure 153. Storage Resource Management Market Share, Windows NT ............................................153
Figure 154. Data Replication ................................................................................................................154
Figure 155. Disaster Recovery Architecture .........................................................................................155
Figure 156. Downtime Costs ................................................................................................................156
Figure 157. Data Replication Projections..............................................................................................156
Figure 158. Replication Market Share, All OS ......................................................................................157
Figure 159. Replication Market Share, UNIX ........................................................................................157
Figure 160. Replication Market Share, Windows NT ............................................................................158
Figure 161. EMC’s Replication Software Can Mirror Data in Two Different Locations...........................159
Figure 162. Traditional DAS (Direct Attached Storage) Backup............................................................161
Figure 163. SAN Backups Can Be LANless .........................................................................................161
Figure 164. SAN Backups Can Be LANless and Serverless.................................................................162
Figure 165. An Example of How HSM Views Data ...............................................................................164
Figure 166. HSM Data Migration — Freeing Primary Storage Capacity by Utilizing Less Expensive
Storage .............................................................................................................................................165
Figure 167. Data Migration in a DAS Architecture ................................................................................166
Figure 168. Data Migration in a SAN Architecture ................................................................................166
Figure 169. HSM and Archive Projections ............................................................................................167
Figure 170. HSM and Archive Market Share, All OS ............................................................................167
Figure 171. HSM and Archive Market Share, UNIX ..............................................................................168
Figure 172. HSM and Archive Market Share, Windows NT ..................................................................168
Figure 173. The Secret Behind a SAN Cloud .......................................................................................172
Figure 174. Mainframe Storage Architecture Using ESCON Directors..................................................174
Figure 175. SAN Networking Component Analysis...............................................................................175
Figure 176. Storage Networking Growth Is Expected to Lead the Industry (YoY Growth Projections)...176
Figure 177. Projected Storage Networking Revenue Growth................................................................176
Figure 178. SAN Market Share by Platform..........................................................................................177
Figure 179. Fibre Channel Host Bus Adapter Growth Projections ........................................................178
Figure 180. HBAs Emerged as a Server-to-Storage I/O Interconnect ...................................................179
Figure 181. HBAs Versus NICs ............................................................................................................180
Figure 182. Host and Target Controllers Negotiate Data Transfer ........................................................181
Figure 183. I/O Interfaces Exist in the Form of Chips, Boards, and Cells..............................................181
Figure 184. Dissection of a Fibre Channel Host Bus Adapter ...............................................................182
Figure 185. HBAs Slide into the PCI Slot, Which Is Connected to the Server or PC’s Backplane (Left) 183
Figure 186. JNI PCI and SBus HBAs ...................................................................................................184
Figure 187. QLogic and JNI Designs Have Similar Data Paths.............................................................185
Figure 188. Emulex LP8000 and Block Diagram Showing Data Path ...................................................186
Figure 189. Historical Growth of Fibre Channel HBA Revenues per Quarter ........................................188
Figure 190. Various HBA Companies’ Positions...................................................................................189
Figure 191. Board/Silicon Relationships ...............................................................................................189
Figure 193. HBA Supplier/Customer Relationships ..............................................................................191

393
The SAN Book III – October 22, 2001

Figure 194. Emulex’s Platform Profile ..................................................................................................192


Figure 195. QLogic’s Platform Profile ...................................................................................................193
Figure 196. JNI’s Platform Profile.........................................................................................................193
Figure 197. Storage Networking Growth (2000–05)..............................................................................194
Figure 198. SAN Switch Diagram..........................................................................................................195
Figure 199. Switch Interconnects .........................................................................................................197
Figure 200. Brocade’s 2Gbps 16-port Switch Architecture....................................................................198
Figure 201. McData’s 2 Gbps 16-port Switch Architecture....................................................................199
Figure 202. QLogic’s 2Gbps 16-port Switch Architecture......................................................................200
Figure 203. Trunking Functionality .......................................................................................................202
Figure 204. Switch Market Growth .......................................................................................................203
Figure 206. Switch Market Share (Revenues) 2000 .............................................................................204
Figure 207. Switch and Director Revenue Growth — Brocade, McData, Inrange, and QLogic..............204
Figure 208. Brocade’s Silkworm Product Family ..................................................................................205
Figure 209. QLogic’s Switch and Director Product Family ....................................................................206
Figure 210. ESCON and FICON Directors ...........................................................................................207
Figure 211. Directors Are Expected to Achieve the Highest Growth in Storage Networking..................208
Figure 212. Director Market Share (Revenues) 2000 ...........................................................................209
Figure 213. Fibre Channel Product Introductions .................................................................................210
Figure 214. Inrange’s Product Portfolio ................................................................................................211
Figure 215. Building a 32-port High Availability Fabric Using 16-port Switches.....................................212
Figure 216. Building a Simple 30 node Fabric Using 16-port Switches .................................................212
Figure 217. Core-to-Edge Storage Networking Architecture .................................................................213
Figure 218. More Likely Core-to-Edge Storage Networking Architecture ..............................................214
Figure 219. McData’s Core-to-Edge Product Portfolio ..........................................................................214
Figure 220. SAN Bridge Diagram .........................................................................................................216
Figure 221. SAN Router Diagram.........................................................................................................217
Figure 222. Crossroads’ Product Portfolio ............................................................................................217
Figure 223. Routers Could Enable Multiple Protocol Connectivity ........................................................218
Figure 224. SAN-to-SAN Interconnectivity via a LAN ...........................................................................219
Figure 225. Nishan’s FC-to-IP Router ..................................................................................................220
Figure 226. SAN Valley’s FC-to-IP Router............................................................................................220
Figure 227. SAN Hub Diagram..............................................................................................................221
Figure 228. Gadzoox’s Product Portfolio ..............................................................................................222
Figure 229. Four-port Hub and Switch Architectures ............................................................................223
Figure 230. Six-port Hub and Switch Architectures (16-port Switches Are More Common) ..................223
Figure 231. Fibre Channel Arbitrated Loop (FC-AL) Diagram ...............................................................224
Figure 232. Various Networking Areas .................................................................................................226
Figure 233. Past and Potential Future Storage Networking Technology Availability..............................228
Figure 234. Evolution of SCSI ..............................................................................................................233
Figure 235. Data Transfer Rates of Various Interface Standards..........................................................234
Figure 236. Parallel Versus Serial Interface .........................................................................................236
Figure 237. SCSI and Fibre Channel Comparisons ..............................................................................237
Figure 238. OSI Reference Model ........................................................................................................241
Figure 239. OSI Versus Ethernet Versus Fibre Channel.......................................................................242
Figure 240. ISO’s OSI Model ...............................................................................................................242
Figure 241. Fibre Channel Versus OSI Model ......................................................................................243
Figure 242. Fibre Channel Five-Layer Model .......................................................................................244
Figure 243. Ethernet Versus OSI Model ...............................................................................................245
Figure 244. Ethernet — Protocol Functions.........................................................................................245
Figure 245. Ethernet Packets ...............................................................................................................246
Figure 246. Ethernet Packet Adds Extension Symbols for Small Data Payloads ..................................246
Figure 247. Fibre Channel Frames.......................................................................................................247
Figure 248. Ethernet, iSCSI, iFCP, Fibre Channel................................................................................250
Figure 249. Newer Protocols Use Elements of Older Ones ..................................................................250
Figure 250. Technologies Have Different Degrees of Backward Compatibility......................................251
Figure 251. FCIP Combines Two Protocols..........................................................................................252
Figure 252. New Technology “Hype” Cycle and Protocol Positioning ...................................................254
Figure 253. Premier InfiniBand Supporters...........................................................................................255
Figure 254. PCI-X Delivers Higher Bandwidth than PCI .......................................................................256
Figure 255. PCI-X Implementation in Next-Generation Servers............................................................257
Figure 256. InfiniBand-Enabled Server Shipments ...............................................................................258
Figure 257. InfiniBand Network ............................................................................................................259
Figure 258. Today’s PCI and PCI-X Server Architectures.....................................................................260
Figure 259. Phase 1: Potential InfiniBand Server Architecture..............................................................260
Figure 260. Phase 2: Potential InfiniBand Server Architecture..............................................................261

394
The SAN Book III – October 22, 2001

Figure 261. InfiniBand Links and Data Throughput Rates.....................................................................261


Figure 262. VI Architecture Comparison...............................................................................................262
Figure 263. VI in the OSI Model ...........................................................................................................263
Figure 264. Potentially RAID and RAIM Architectures Could Look Very Similar ...................................264
Figure 265. Networking Diagram..........................................................................................................265
Figure 266. IP SAN Router...................................................................................................................267
Figure 267. SAN and NAS Convergence with EMC’s Celerra...............................................................270
Figure 268. SAN, NAS, and DAS (Direct Attached Storage) Convergence with EMC’s Celerra............270
Figure 269. Total Storage Convergence Including Remote Mirroring....................................................271
Figure 270. 100x4 Quad-Port Server Accelerator.................................................................................285
Figure 271. 100x1 Single-Port Server Accelerator................................................................................285
Figure 272. Ancot — FCAccess 2000 Fibre Channel Analyzer.............................................................286
Figure 273. Ancot — The Ultra2160 SCSI Bus Analyzer ......................................................................286
Figure 274. ATTO Technology’s Enterprise Solutions ..........................................................................289
Figure 275. ATTO ExpressPCI FC 3305 ..............................................................................................290
Figure 276. ATTO FibreBridge 3200R..................................................................................................290
Figure 277. ATTO FibreCenter, Fibre Channel hub ..............................................................................290
Figure 278. ATTO FibreCenter 2100R .................................................................................................291
Figure 279. Baydel’s O/ESR 2000........................................................................................................293
Figure 280. Baydel’s O/ESR RAIDER-5 ...............................................................................................293
Figure 281. An Example RAIDMON Software Screen. Error Notification: Failed Disk .........................294
Figure 282. BlueArc’s Si7500 ...............................................................................................................295
Figure 283. Bus-Tech’s Original DataBlaster........................................................................................297
Figure 284. Bus-Tech’s DataBlaster 2..................................................................................................297
Figure 285. Bus-Tech’s NetShuttle.......................................................................................................297
Figure 286. Bus-Tech’s NetShuttle Product Features...........................................................................298
Figure 287. Chaparral Network Storage — FS2620: 2Gbps Fibre Channel-to-Ultra160 SCSI Storage
Router...............................................................................................................................................300
Figure 288. Chaparral Network Storage — G6322: Ultra 160-to-Ultra 160 SCSI .................................300
Figure 289. CMD Technology — CMD Titan CRA-7280 RAID controller ..............................................301
Figure 290. The Datacore Software Drag-and-Drop Interface...............................................................304
Figure 291. DataDirect Networks’ SAN DataDirector............................................................................305
Figure 292. Multiplatform Environments ...............................................................................................306
Figure 293. DataDirect Networks’ OEM Exclusive EV-5000, Fastest Fibre Channel Network RAID
System — 185MB/Sec Sustained Throughput...................................................................................307
Figure 294. DataDirect Networks’ EF-2000 Fibre Channel SAN Ready RAID Solution .........................307
Figure 295. Hitachi Data Systems Solutions.........................................................................................310
Figure 296. Hitachi Hi-Star Switched Architecture ................................................................................311
Figure 297. I-TECH — IFC-4 Fibre Channel Tester..............................................................................313
Figure 298. I-TECH —Satellite IFC-4016 Fibre Channel Analyzer........................................................313
Figure 299. Lane15 Software’s Product Architecture............................................................................317
Figure 300. Mellanox Technologies — Two-Port Card .........................................................................319
Figure 301. Nishan Systems’ SoIP Product Family...............................................................................321
Figure 302. Nishan Systems’ IPS (IP Storage) 1000 Gateway .............................................................321
Figure 303. Nishan Systems’ IPS 2000 Switch.....................................................................................322
Figure 304. Nishan Systems’ IPS 3000 Switch.....................................................................................322
Figure 305. Nishan Systems’ SANvergence Management Suite ..........................................................322
Figure 306. NSI Software’s Double-Take .............................................................................................323
Figure 307. The NexStor 3250S — Dual RAID Controller SCSI Storage System .................................325
Figure 308. The NexStor 1202S — JBOD SCSI Storage System .........................................................325
Figure 309. AdminiStor Storage Management Software .......................................................................326
Figure 310. OmegaBand’s IBgate 1000 ...............................................................................................326
Figure 311. The Pirus Storage Utility Switch ........................................................................................328
Figure 312. PolyServe’s SANs Vision...................................................................................................329
Figure 313. PolyServe Matrix Server....................................................................................................330
Figure 314. Power Quest Desktop Solutions ........................................................................................331
Figure 315. PowerManage Suite ..........................................................................................................332
Figure 316. Prisa’s VisualSAN Network Manager.................................................................................333
Figure 317. Prisa’s VisualSAN Performance Manager .........................................................................334
Figure 318. Raidtec’s Products ............................................................................................................335
Figure 319. Focus Technologies ..........................................................................................................336
Figure 320. SANgate’s Enterprise Storage Appliance (ESA) ................................................................338
Figure 321. Storagetone.com Portal.....................................................................................................340
Figure 322. Storagetone Fibrecloud .....................................................................................................340
Figure 323. SL1000 Gateway...............................................................................................................341
Figure 324. SVS Management System ................................................................................................342

395
The SAN Book III – October 22, 2001

Figure 325. XIOtech Product Family.....................................................................................................344


Figure 326. Storage Virtualization Manager (SVM)...............................................................................348
Figure 327. Tantia NSM Enterprise-class Backup and Recovery Solution ............................................349
Figure 328. Tantia Harbor Backup .......................................................................................................349
Figure 329. Tantia Enterprise Agent Suite............................................................................................350
Figure 330. Tantia Technologies — RS/6000 Solution .........................................................................350
Figure 331. Tantia Technologies — Bus-Tech Datablaster 2 Solution ..................................................350
Figure 332. Tantia Technologies — Crossroads PCI ESCON Solution.................................................350
Figure 333. Storage Monitoring Manager (Tantia SMM) .......................................................................351
Figure 334. The Troika Networks Solution............................................................................................353
Figure 335. Troika Networks’ Zentai Controller.....................................................................................354
Figure 336. Paladin C-series ................................................................................................................357
Figure 337. Vicom Systems’ Independent Distributed Routing .............................................................359
Figure 338. Vicom Systems’ SV Routers..............................................................................................359
Figure 339. Vicom Systems’ SV Bridges ..............................................................................................359
Figure 340. VIEO Fabric Manager: Core Services — WAN/MAN Solution...........................................361
Figure 341. YottaYotta’s Scalable Solution...........................................................................................363
Figure 342. YottaYotta’s WAN/MAN Solution .......................................................................................364

396
The SAN Book III – October 22, 2001

Notes

397
The SAN Book III – October 22, 2001

Notes

398
The SAN Book III – October 22, 2001

Companies Mentioned in this Report


3COM (COMS-$4.32; NR) Inrange# (INRG-$7.17; 1H)
Adaptec (ADPT-$10.04; NR) Intel (INTC-$24.38; 1M)
Advanced Digital# (ADIC-$12.10; 3H) Interphase (INPH-$4.11; NR)
Agilent# (A-$23.23; 2M) Intevac# (IVAC-$3.75; 3S)
AOL# (AOL-$33.50; 1H) Intuit (INTU-$35.22; 2H)
Apple (AAPL-$17.99; 3H) Iomega (IOM-$6.72; 3S)
Arc International (ARK.L-£38.50; NR) iVillage (IVIL-$0.96; NR)
AT&T# (T-$19.20; 3M) JNI# (JNIC-$7.45; 3H)
Auspex (ASPX-$2.34; NR) Legato (LGTO-$8.60; 3S)
Bakbone (BKB.TO-C$1.70; NR) LSI Logic (LSI-$15.80; 3S)
Bell Microproducts (BELM-$8.93; 2H) Lucent# (LU-$6.83; 2H)
BMC Software (BMC-$15.53; 1H) Maxtor# (MXO-$4.50; 2H)
British Telecom (BTY-$48.60; NR) McData (MCDT-$14.65; NR)
Brocade (BRCD-$24.17; 1H) Microsoft (MSFT-$58.06; 2H)
Cisco (CSCO-$16.21; 1H) Motorola# (MOT-$17.58; 2M)
Citigroup (C-$44.80; NR) NEC (NIPNY-$9.30; 2H)
Compaq# (CPQ-$9.80; 1H) Network Appliance (NTAP-$11.61; 3H)
Computer Associates (CA-$29.51; NR) Nortel (NT-$5.94; 2H)
Computer Network Technologies (CMNT-$15.10; NR) Novell (NOVL-$3.85; NR)
Crossroads (CRDS-$2.55; NR) ONI Systems# (ONIS-$7.07; 1S)
Dell (DELL-$23.58; 3H) OTG Software# (OTGS-$5.50; 2H)
Deutsche Telekom (DT-$15.95; 3H) Procom (PRCM-$2.03; NR)
Dot Hill Systems (HIL-$1.58; 3S) QLogic (QLGC-$36.12; 3H)
Eastman Kodak# (EK-$34.99; 3M) Quantum (DSS-$8.55; 3H)
EMC (EMC-$13.10; 2M) Read-Rite (RDRT-$4.48; 2S)
Emulex (EMLX-$20.76; 3S) Samsung (05930.KS-163,000SKW; 1H)
Falconstor (FALC-$9.10; NR) SGI (SGI-$0.45; 4S)
Fujitsu (6702.JP0-¥997; NR) Storage Technology (STK-$14.80; 3H)
Gadzoox (ZOOX-$0.93; NR) Sun Microsystems# (SUNW-$9.68; 1H)
GE (GE-$38.86; 1L) Symantec (SYMC-$51.04; NR)
Hewlett-Packard# (HWP-$18.12; 1M) United Online (UNTD-$2.79; NR)
Hitachi (HIT-$73.40; 3M) Veritas (VRTS-$27.93; 2H)
Hutchinson Technology# (HTCH-$19.60; 3S) Western Digital# (WDC-$2.54; 3H)
IBM# (IBM-$102.00; 1M)

399
The SAN Book III – October 22, 2001

ADDITIONAL INFORMATION IS AVAILABLE UPON REQUEST US09J159


# Within the past three years, Salomon Smith Barney, including its parent, subsidiaries, and/or affiliates, has acted as manager or co-manager of a public offering of the securities
of this company.
Salomon Smith Barney (“SSB”), including its parent, subsidiaries, and/or affiliates (“the Firm”), usually makes a market in the U.S.-traded over the counter securities
recommended in this report and may sell to or buy from customers, as principal, securities recommended in this report. The Firm or employees preparing this report may have a
position in securities or options of any company recommended in this report. An employee of the Firm may be a director of a company recommended in this report. The Firm may
perform or solicit investment banking or other services from any company recommended in this report.
Securities recommended, offered, or sold by SSB: (i) are not insured by the Federal Deposit Insurance Corporation; (ii) are not deposits or other obligations of any insured deposit-
ory institution (including Citibank); and (iii) are subject to investment risks, including the possible loss of the principal amount invested. Although information has been obtained
from and is based upon sources SSB believes to be reliable, we do not guarantee its accuracy and it may be incomplete or condensed. All opinions and estimates constitute SSB’s
judgment as of the date of the report and are subject to change without notice. This report is for informational purposes only and is not intended as an offer or solicitation for the
purchase or sale of a security.
This report has been approved for distribution in the United Kingdom by Salomon Brothers International Limited, which is regulated by the Securities and Futures Authority. The
investments and services contained herein are not available to private customers in the UK and South Africa. This report was prepared by SSB and, if distributed in Japan by Nikko
Salomon Smith Barney Limited, is being so distributed under license. This report is made available in Australia through Salomon Smith Barney Australia Securities Pty Ltd. (ABN 64
003 114 832), a Licensed Securities Dealer, and in New Zealand through Salomon Smith Barney New Zealand Limited, a member firm of the New Zealand Stock Exchange. This
report does not take into account the investment objectives or financial situation of any particular person. Investors should obtain advice based on their own individual
circumstances before making an investment decision. Salomon Smith Barney Securities (Proprietary) Limited is incorporated in the Republic of South Africa (company registration
number 2000/025866/07) and its registered office is at Grosvenor Corner, 195 Jan Smuts Avenue, Rosebank, 2198, Republic of South Africa.
Guide to investment ratings: RANK is a guide to the expected total return over the next 12-18 months. The total return required for a given rank depends on the degree of risk (see
below) in a stock. The higher the risk, the higher the required return. For example, a 1 (Buy) rating indicates a total return ranging from 15% or greater for a low-risk stock to 30%
or greater for speculative stocks. Estimated returns for other risk categories are scaled accordingly. RISK takes into account predictability of earnings and dividends, financial
leverage, and stock price volatility. L (Low Risk): predictable earnings and dividends, suitable for conservative investors. M (Medium Risk): moderately predictable earnings and
dividends, suitable for average equity investors. H (High Risk): earnings and dividends are less predictable, suitable for aggressive investors. S (Speculative): very low
predictability of fundamentals and a high degree of volatility, suitable only for investors/traders with diversified portfolios that can withstand material losses. V (Venture): indicates
a stock with venture capital characteristics that is suitable for sophisticated investors with a high tolerance for risk and broadly diversified investment portfolios. A thorough
explanation of the ratings system is available upon request.
Investing in non-U.S. securities, including ADRs, by U.S. persons may entail certain risks. The securities of non-U.S. issuers may not be registered with, or be subject to, the
reporting requirements of, the U.S. Securities and Exchange Commission. There may be limited information available on foreign securities. Foreign companies are generally not
subject to uniform audit and reporting standards, practices and requirements comparable to those in the U.S. Securities of some foreign companies may be less liquid and their
prices more volatile than securities of comparable U.S. companies. In addition, exchange rate movements may have an adverse effect on the value of an investment in a foreign
stock and its corresponding dividend payment for U.S. investors. Investors who have received this report from the Firm may be prohibited in certain U.S. states from purchasing
securities mentioned in this report from Salomon Smith Barney. Please ask your Financial Consultant for additional details.
Salomon Smith Barney is a service mark of Salomon Smith Barney Inc.
© Salomon Smith Barney Inc., 2001. All rights reserved. Any unauthorized use, duplication, or disclosure is prohibited by law and will result in prosecution.

400

Potrebbero piacerti anche