Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm

Energy and Buildings 43 (2011) 21452155

Contents lists available at ScienceDirect
Energy and Buildings
j our nal homepage: www. el sevi er . com/ l ocat e/ enbui l d
Evaluation of air management systems thermal performance for superior
cooling efciency in high-density data centers
Jinkyun Cho
a
, Byungseon Sean Kim
b,
a
Construction Technology Center, Samsung C&T Corporation, Seoul 137-070, South Korea
b
Department of Architectural Engineering, Yonsei University, Seoul 120-749, South Korea
a r t i c l e i n f o
Article history:
Received 5 September 2010
Received in revised form12 March 2011
Accepted 26 April 2011
Keywords:
Data center
Aisle partition system
Cooling efciency
Re-circulation air
By-pass air
CFD modeling
a b s t r a c t
In a typical data center, large numbers of IT sever racks are arranged in multiple rows. In IT environments,
in which extensive electronic hardware is air-cooled, cooling system inefciencies result when heated
exhaust air from equipment prematurely mixes with chilled coolant air before it is used for cooling. Mix-
ing of chilled air before it is used with heated exhaust air results in signicant cooling inefciencies in
many systems. Over-temperatures may not only harm expensive electronic equipment but also interrupt
critical and revenue generating services. The cool shield is a cost effective aisle partition system used to
contain the air in cold aisles and hot aisles of an IT server room. This paper focuses on the use of perfor-
mance metrics for analyzing a vertical aisle partition system in high-density data centers. Performance
metrics provide a great opportunity for the data center industry at large. They could form the foundation
for a standardized way of specifying and reporting various cooling solutions. The Rack Cooling Index
(RCI) is a measure of how well the system cools the IT servers within the manufacturers specications,
and the Return Temperature Index (RTI) is a measure of the energy performance of the air-management
system. Combined, they provide an opportunity to judge the performance of the aisle partition system
in an objective way subsequent to comprehensive CFD modeling.
2011 Elsevier B.V. All rights reserved.
1. Introduction
A data center is a centralized repository, either physical or
virtual, for the storage, management, and dissemination of data
and information organized around a particular body of knowledge
or pertaining to a particular business. The air-conditioning sys-
tem in todays data center must be capable of supporting on a
24/7, 365 days/year [1]. With the rapid growth of the information-
basedindustry, data centers have become a prevalent phenomenon
in both the public and private sectors. Such facilities are widely
used for web-hosting, central depository information bases of gov-
ernmental organizations and research units, intranet, nancial
transaction processing, and other activities [2]. Data centers are
facilities that house IT servers and data storage systems. To ensure
that these computer systems function reliably, they must be ade-
quately cooled. Each IT server must receive a certain minimum
amount of cooling air, determined by its heat generation rate. Thus,
the key to guarantee equipment reliability is to ensure that the
cooling air distributes properly throughout the data center, that
is, that the supplied airow meets the airow demand at each

Corresponding author. Tel.: +82 2 2123 2791; fax: +82 2 365 4668.
E-mail address: sean@yonsei.ac.kr (B.S. Kim).
location [3]. Average power densities in data centers are rapidly
increasing and are expected to reach up to 3000W/m
2
in the next
5 years [4]. During the last few years, data centers have increas-
ingly drawn much concern fromutilities, building owners, facility
managers, and IT equipment manufacturers. This can be attributed
to some of the ndings as follows; at these heat dissipation levels,
the inefciencies of state-of-the-art data centers are intensifying.
These inefciencies include re-circulation of warmair into the cold
aisle, short-circuiting (or by-pass) of the cold air back into the com-
puter roomair conditioning (CRAC) unit, lack of information about
local conditions which could lead to hot spots, mal-provisioning
of cooling resources that leads to inefcient systemoperation, and
inefcient workload placement [5,6].
Thermal management of IT equipment relies heavily on how
well cool air is distributed in the equipment room. Data centers
need a tool for evaluating and designing the thermal environment
to ensure effective equipment cooling without excessive energy
usage [7]. Telcordia [8] was developed to update historical envi-
ronmental criteria in telecom facilities. It introduced a common
language and new concepts, requirements, and objectives for
designing and operating equipment facilities. Data centers thermal
guidelines produced by ASHRAE Technical Committee 9.9 [9]
adopted several of the ideas that were introduced by Telcordia
[8]. The document provides equipment manufacturers and facil-
ity operations personnel with a common set of guidelines for
0378-7788/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2011.04.025
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
2146 J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155
Nomenclature
Q total dissipation fromdata center components (W)
C
p
specic heat of air at constant pressure (Pa)
m mass owrate of air through a rack (CMH)
T temperature (

C)
n total number of intakes
x intake x
Superscripts
r rack
c CRAC
Subscripts
in inlet
out outlet
ref CRAC supply
i,j Cartesian direction
max-rec maximum recommended/some guideline or stan-
dard
max-all maximumallowable/some guideline or standard
min-rec minimum recommended/some guideline or stan-
dard
min-all minimumallowable/some guideline or standard
Return return air (weighted average)
Supply supply air (weighted average)
Equip rise across the electronic equipment (weighted
average)
environmental conditions in data centers. Sharma et al. [10] intro-
duced two dimensionless parameters as the Supply Heat Index
(SHI) and the Return Heat Index (RHI) for evaluation of the thermal
performance of data centers. These indices provide a way of under-
standing the convective heat transfer in equipment rooms with
raised oors. Energy efciency can be impacted not only by inad-
equate cooling systems but also equipment room congurations
that allowhot and cold air to mix. Herrlin [11] studied rack cooling
effectiveness in data centers. The Rack Cooling Index (RCI) is a
measure of howeffectively equipment racks are cooled and main-
tained within industry temperature guidelines and standards. It is
also well suited as a design specication for new data centers. An
example with under-oor versus overhead cooling is included in
the reference above to demonstrate the use of the index. VanGilder
and Shrivastava [12] introduced the dimensionless Capture Index
(CI), which is a cooling performance metric at the equipment rack
level. The CI is also typically computed by CFD modeling.
Cho et al. [13] veried through CFD simulation the thermal
performances of 6 types of air distribution systems generally appli-
cable for high-density data centers. But only a relative evaluation
based on the temperature and airow distribution of the IT server
room was performed without using evaluation indices. Of the
various systems, the under-oor system showed the most stable
indoor environment distribution, but even in this system, the air
re-circulation or air bypass continued to occur.
Herrlin [14,15] carried out a study on the Return Temperature
Index (RTI) which is a measure of the net level of by-pass air or
net level of recirculation air in the data center. Both effects are
detrimental to the overall energy and thermal performance of the
space.
The main task for a data center facility is to provide an adequate
equipment environment, and a relevant metric for equipment
intake temperatures should be used to gauge the thermal envi-
ronment. RCI is a measure of how effectively equipment racks
are cooled and maintained within industry temperature guide-
lines and standards [11]. RTI is a measure of the level of by-pass
air or re-circulation air in the equipment room. By-pass air does
not contribute to the cooling of the electronic equipment, and it
depresses the return air temperature. Recirculation, on the other
hand, is one of the main reasons for hot spots or areas signicantly
hotter than the ambient temperature [14]. A successful design of
the air distribution system(ADS) requires an understanding of the
fundamental mechanisms governing the distribution of chilled air
in data centers. The aim of this study is to propose the vertical
aisle partition system for increasing the cooling efciency related
to the IT server heat removal in a data center and objectively ana-
lyzing its performance. This paper proposed additional measures
to improve the thermal performance of the under-oor system,
whichis most ubiquitously appliedincurrent data centers, andver-
iedthe results byutilizingquantitative evaluationindices. Anaisle
partition systemis important for adequately cooling the IT equip-
ment and controlling the associated energy costs. Computational
FluidDynamics (CFD) modelinghas the capacitytohelpunderstand
how a cooling solution will performprior to being built. However,
modeling also has the capacity to generate an unwieldy amount of
data. The crux of the matter is to know what to look for and then
objectively characterize and report the performance.
2. Data centers thermal management
2.1. Importance of thermal management in the server rooms
Fromthe composition of the server roomwhich is the core ele-
ment in a data center, as shown in Fig. 1, it can be seen that the IT
server that processes and saves data takes up the largest portion,
and the other components are the data cables that can be intercon-
nected, the power cables and the rack for the server installation.
In general data centers, cooling designs utilize CRAC units along
the perimeter of the room and perforated tiles in the raised oor
these designs are not capable of cooling higher density cabinet
loads. Since the main function of the data center ADS is to prevent
overheating of IT equipment, it is important to make sure the air
is not intermixed at the air inlet and outlet installed on the equip-
ment. Many data center operators take a relaxed attitude toward
by-pass air, which is produced by the CRACs and returns directly
without cooling servers. By-pass air causes a gap between the tem-
perature of air produced by the CRAC units and the temperature
of the air at the inlet to the server. CRAC units are designed to
produce lower temperatures for the sole reasonof offsetting the re-
circulationof air fromthe outlet of the server to its inlet. Air by-pass
also starves the servers of air, which increases re-circulation. The
goal of air management is to minimize re-circulation of hot air and
minimize by-pass of cold air in the data center. Successfully imple-
mented, both measures result in energy savings and better thermal
conditions. Furthermore, improved air management is essential to
maximizing data center free cooling opportunities. When the IT
server rooms airows areefcientlymaintained, thegreatest effect
can be obtained with the smallest effort. However, because there
is still a lack of understanding on this matter, most data centers
fail to seize the opportunity of improving cooling efciency and
energy performance, and despite having enough cooling capacity,
they install additional cooling equipment to resolve the problem
of local temperature rise. The server room air management basi-
cally involves all the air supplied by the CRAC unit owing into
the IT server, and after the heat removal, it involves the return of
hot air to the CRAC unit. However, in reality, factors that obstruct
airstreams, such as air re-circulation, or by-pass, as shown in Fig. 2,
lower the cooling efciency and a vicious cycle of rising local tem-
perature occurs [16]. By measuring the temperature at the front
and rear of an IT server, Cho et al. [17] examined the distribution
of air temperature after heat is actually removed from the server.
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155 2147
Fig. 1. A typical IT server roomand data center infrastructures.
Source: Modied fromhttp://www.oriensoft.com/Infrastructure.
The IT server room has an under-oor distribution system where
the lower part of the server front is near the supply air diffuser
and thus able to maintain a constant temperature, whereas, at the
upper part, the re-circulating air and indoor air become mixed and
consequently severe temperature changes occur [17]. Such a phe-
nomenon occurs mainly at the upper part of the server, and 65% of
the total system impairments occur at the upper third of the rack
server, resulting in considerably large economic losses fromserver
malfunctions and breakdowns [18].
2.2. Air distribution systems for efcient data center cooling
The former phase of the study has been developed by means of
CFDtechniques. Cho et al. [13] analyzed the impact of cold aisle and
hot aisle separation on cooling efciency, presented an alternative
to the ADS generally applied in data centers, and reviewed the air
temperature and velocity distribution of each system. Efciency
evaluation for each ADS was carried out on six alternative cases
basedonthe methodof providing supply air andreturnair. Fromits
average temperature distribution, under-oor distribution-locally
ducted supply/locally ducted return infrastructures (U-LS/LR) were
veried to be the most efcient air distribution method if the slight
air re-circulation at the upper part of the server is effectively pre-
vented [13]. Airow and cooling performance of data centers has
seen the set up of performance metrics. SHI is a function of the rack
inlet, rack outlet, and CRAC outlet temperatures. This parameter
signies a ratio of the heat gained by cold aisle air before it enters
the racks to the heat gained by the air leaving the racks. The numer-
ator in the RHI parameter signies the effective heat dissipation of
the data center. The utilization of dimensionless parameters allows
these formulas to be scalable for any size system. These equations
are dened as:
SHI =
_
Q
Q +Q
_
(1)
RHI =
_
Q
Q +Q
_
(2)
Q =

i
m
r
i,j
C
p
((T
r
out
)
i,j
(T
r
in
)
i,j
) (3)
Q =

i
m
r
i,j
C
p
((T
r
in
)
i,j
T
ref
) (4)
SHI +RHI = 1 (5)
Consider Q as the total heat dissipation from all the racks in
the data center and Q as the rise in enthalpy of the cold air before
entering the racks. Where m
r
i,j
is the mass owof air throughthe ith
rack in the jth rowof racks, (T
r
in
)
i,j
and (T
r
out
)
i,j
are the average inlet
and outlet temperature fromthe ith rack in the jth rowof racks. T
ref
denotes the vent tile inlet air temperature, assumed to be identical
for all rows. Assuming no heat transfer in the plenum, the vent tile
air temperature and CRAC supply air temperature are considered
to be equal and referred to as reference temperatures for enthalpy
calculations [10].
In terms of application, if SHI is a high value, it indicates that the
inlet temperatures are high, which is likely due to re-circulation
or mixing prior to entering the rack. SHI serves as a good indica-
tion of energy efciency of a system, as well as an indication of
failure or reliability problems due to high systemtemperatures. A
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
2148 J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155
Fig. 2. Out-of-control airowin the server rooms. (a) Re-circulation airowand (b)
by-pass airow.
low RHI is an indication of rack exhaust heat mixing with cold air
inside the aisle, ceiling, or between the rack and walls. Evaluation
of cooling efciency was carried out by using the CFD simulation
results of the six ADSs. On the basis of the total air volume (m)
supplied to the IT server room and the supply air temperature
(T
ref
), the average cold aisle temperature calculated for each sys-
tem was assumed as the servers inlet air temperature (T
r
in
) and
the average hot aisle temperature as the servers outlet air tem-
Fig. 3. Sumof heat indices for different infrastructures.
perature (T
r
out
), and substituting these in Eqs. (3) and (4), SHI and
RHI were calculated. The SHI is a measure of the amount of heat
that inltrates the cold aisle and is subsequently pulled into the
heat-producing equipment. Heat inltration into cold aisles is ulti-
mately the result of re-circulation patterns in the room. The power
density in the data center was 4kW/rack. As shown in Table 1, the
average hot aisle temperatures obtained fromALT-1 to ALT-6 were
26.0

C, 24.9

C, 27.1

C, 28.2

C, 29.6

C and 29.3

C, respectively.
For the CRAC ooded supply infrastructures (ALT-1/ALT-2), the SHI
and RHI values were 0.59/0.72 and 0.41/0.28, respectively, and
for overhead locally ducted supply infrastructures (ALT-3/ALT-4),
the SHI and RHI values were 0.62/0.29 and 0.38/0.71, respectively.
The SHI and RHI values for the under-oor locally ducted supply
cases were 0.37/0.28 and 0.63/0.72, respectively. Fig. 3 shows the
stacked column chart with SHI and RHI for room-return and ceil-
ing return infrastructures. A numerically signicant increase in SHI
and RHI of 250% compared with ALT-2 was noted and the under-
oor distribution-locally ducted supply and return infrastructure
(U-LS/LR) was the most efcient ADS in high compute density data
centers except fully ducted supply and return infrastructures.
Table 1
RHI and SHI values versus supply and return infrastructures for a data center feature.
Air distribution
system(ADS)
ALT-1 (O-CS/CR)
a
ALT-2 (O-CS/LR)
b
ALT-3 (O-LS/CR)
c
ALT-4 (O-LS/LR)
d
ALT-5 (U-LS/CR)
e
ALT-6 (U-LS/LR)
f
Remark
Environment Hard oor Hard oor Hard oor Hard oor Raised oor Raised oor Floor type
Supply air CRAC ooded CRAC ooded Locally ducted
(Overhead)
Locally ducted
(Overhead)
Locally ducted
(Under-oor)
Locally ducted
(Under-oor)
Return air CRAC ooded Locally ducted CRAC ooded Locally ducted CRAC ooded Locally ducted
m Supply air volume:
45,600 CMH
CRAC [12]
T
ref
Supply air
temperature: 13

C
CRAC [12]
T
r
in
20.7

C 21.6

C 21.7

C 17.5

C 19.1

C 17.5

C Cold aisle [12]

T
r
out
26.0

C 24.9

C 27.1

C 28.2

C 29.6

C 29.3

C Hot aisle [12]

SHI 0.59 0.72 0.62 0.29 0.37 0.28 Target: 0.0
RHI 0.41 0.28 0.38 0.71 0.63 0.72 Target: 1.0
Source: Modied result data fromCho et al. [13].
a
O-CS/CR: overhead distribution-CRAC ooded supply and CRAC ooded return.
b
O-CS/LR: overhead distribution-CRAC ooded supply and fully ducted return.
c
O-LS/CR: overhead distribution-locally ducted supply and CRAC ooded return.
d
O-LS/LR: overhead distribution-locally ducted supply and locally ducted return.
e
U-LS/CR: under-oor distribution-locally ducted supply and CARC ooded return.
f
U-LS/LR: under-oor distribution-locally ducted supply and locally ducted return.
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155 2149
Fig. 4. Air distribution systemapplications in high compute density data centers. (a) Typical under-oor air cooling systemconguration, (b) typical conguration with aisle
partition system, and (c) typical conguration with aisle enclosure system.
3. Cooling and airowperformance improvement
3.1. Design of IT server room: cold aisle and hot aisle separation
As shown by the results in the previous section, prevention of
the air from the cold aisle and the hot aisle intermixing; air re-
circulation and by-pass must be considered in order to prevent a
drop in the rack cooling efciency. However, there is a limitation
in effectively preventing such phenomenon in an open space IT
server roomwith only the location of supply and return air infras-
tructures. The reality is that the problemof rising temperature (hot
spot) continues to occur in currently operating data centers (see
Fig. 4(a)). Therefore, there is a need to improve air distribution ef-
ciency through additional physical barrier installation, and cooling
efciency can be improved by installing a simple partition wall on
the rack server. Two types of formation are possible here: the aisle
partition systemvertically dividing the cold aisle and the hot aisle,
as shown in Fig. 4(b): and the aisle enclosure systemthat blocks off
the upper part of the cold aisle, as shown in Fig. 4(c). In terms of
cooling efciency, the aisle enclosure systemthat completely sur-
rounds the cold aisle can more effectively prevent air re-circulation
in comparison with the aisle partition system that simply forms a
vertical wall. However, because the aisle enclosure systemcan act
as an obstructing factor when relocating the IT server, setting up
re suppression systems [19] as clean agent systems and water-
based systems (gas and sprinkler heads installation location), and
responding quickly to a re, the installation of the aisle partition
systemis more reasonable. In addition, since a large data center has
a large amount of IT equipment, the matter of initial cost cannot be
overlooked.
3.2. CFD modeling
In this section, realistic alternatives will demonstrate how CFD
simulation can be used to estimate the rack intake temperatures
in a conventional data center with two scenarios: when the ver-
tical aisle partition wall is applied on the basis of the under-oor
distribution-locally ducted supply and locally ducted return sys-
tem(U-LS/LR) widely used in data centers, as veried in preceding
studies as the most effective ADS, and when it is not applied. The
computational study reported here is conducted using the com-
mercial software package STAR-CD; a customized CFD package
designed specically for airow distribution and thermal envi-
ronment in data centers. The ke model is the most appropriate
for large, open space environments because of the way it calcu-
lates the turbulent viscosity andconductivity. CFDmodeling allows
three-dimensional analyses of key environmental variables such
as temperature and airow. The two ADSs are applied in a typ-
ical data center with hot and cold aisles (Fig. 5). The 22,500m
2
of raised oor space in the data center accommodated an aver-
age of 4kWrack load. In the model, the numerical computational
domain was part of an IT server room. The domain area was mod-
eled after a four rows module (8.4m15m) and the ceiling is
3.0m above the raised oor. The hot aisles (0.9m) and cold aisles
(1.2m) are sized per ASHRAE recommendations [19]. The conven-
tional U-LS/LR ADS (Fig. 6(a)) is based on an access oor with
(600mm600mm) perforated oor tiles. Air is generally supplied
by down-ow CRAC units. The tile airow is uniformassuming an
even pressure distribution in the under-oor plenum; the actual
plenum airow and pressure distribution were not modeled. The
supply temperature is 16

C. The aisle partition system (Fig. 6(b))

consists of vertical partition walls installed above server racks. The
IT servers and supply air inlets and return air outlets of CRAC units
were arrayed across the room to form the room model. Fig. 7(a)
shows the overall model of the room. 300,000 grid cells were
arrayed across the solution domain. For modeling purposes, each
2.1m IT equipment rack is subdivided into 4 shelves, each hav-
ing four monitor points to establish the mean temperature at the
air intake [9]. Each shelf is individually fan cooled; the equipment
has front-to-rear cooling withanequipment-cooling class F-R [8].
The temperature rise across the shelves is over 15

C. As shown
in Fig. 7(b), the racks were modeled as enclosures with an insert
rectangular block. The block had four sets of re-circulating open-
ings. Each part opening pair was assigned a ow of 160CMH and
the total airow rate of the block was about 650CMH. All pairs
were assigned a heat load of 1000Wsuch that four compartments
within an IT server (rack) were at full power, and each rack was dis-
sipating 4kW. The boundary conditions of each case are given in
Table 2.
3.3. Simulation results
CFD simulation allows visualization of environmental param-
eters throughout the IT server room, including air temperature.
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
2150 J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155
Fig. 5. Modeled data center with equipment racks and hot and cold aisles. (a) Plan and (b) section.
Fig. 8 depicts the temperature distributions for the x, y and z cross-
sections inFig. 7(a). Theperspectiveis a viewtowardthecoldaisles;
the gray rectangles are the IT server racks with 4 shelves each. Dif-
ferent temperatures are represented by different colors, fromblue
(cold) to red (hot). (For interpretation of the references to color in
this text, the reader is referred to the web version of this article.)
It is clear that the two cooling systems result in different temper-
ature conditions in the cold aisle. For the typical under-oor air
cooling system, belowa height of 1.8m, as shown in Fig. 8(a), very
good airow is shown. However, re-circulation occurs at the top
of the racks due to an insufcient supply owrate within z =2.1m
and 2.7m. IT server shelves above the interface are exposed to sig-
nicant over-temperatures. Therefore, it is possible to block the
airowand enable smooth outowthrough the return air grille by
installing a partition that vertically divides the cold aisle and the
hot aisle to a 0.6m level of the top of the server. The aisle parti-
tion systemproduces a well-mixed cold aisle and the servers draw
air with even temperature conditions. The y-axis cold aisle vertical
temperature distribution shown in section in Fig. 8(a) shows that,
up to 1.8m, the temperature distribution is maintained at a rela-
tively stable level of about 1820

C, but in the upper part of the

server (1.82.1m), because of air re-circulation fromthe hot aisle,
about 30

C is maintained, even though it is a cold aisle section. If

the rack is seen as vertically differentiated into 4 parts, when such
a high temperature is maintained for the highest part of the server,
it becomes the main cause of equipment malfunction. If the aisle
partition system is installed on the cold aisle basis (Fig. 8(b)), the
air with a temperature of about 16

C is supplied from the oor

throughout the cold aisle, and although the temperature rises as
the air ows to the upper part, the air temperature at 30

C or
higher fromthe hot aisle re-circulation and mixing in the cold aisle
is observed to be greatly reduced. When the entire x-axis cold aisle
was examined, depending on whether the aisle partition walls are
installed, the appearance of a substantial difference in the IT server
inlet temperature distribution can be seen. That is, the difference of
highandlowtemperaturebyITserver height is less than10

Cwhen
the partitionwalls are installed, but whenthey are not installed, the
difference can increase to a maximumof 15

C. Furthermore, in the
hot aisle, there is a difference in the temperature of the air nally
back to the CRAC units, depending on whether there is chilled air
re-circulation. Depending on whether aisle partitions are installed,
air re-circulation can increase the average temperature at the loca-
tion of the IT server rooms return air grille by about 1.4

C, from
31.8

C to 33.2

C.
Fig. 6. Selected cooling systems. (a) SystemA: typical under-oor air cooling systemand (b) systemB: under-oor air cooling with aisle partition system.
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155 2151
Fig. 7. Basic case layout for IT server room air distribution system infrastructure.
(a) CFD model construction and (b) simplied denition of a server rack.
4. Rack cooling performance of data centers
4.1. Cooling performance metrics: RCI and RTI
Generally, the indices that manage data centers focus on the
temperature of IT servers. In recent related research, methods for
evaluating cooling efciency using indices such as SHI, RHI, RCI
and RTI have been presented [11,14,15]. These measurements are
used for preventing the air with increased temperature after heat
removal from re-circulating back to the IT server and the chilled
air without the heat properly removed fromby-pass back into the
CRAC unit. The SHI and RHI (reviewed in Section 2) are dimension-
less indices that show the volume of mixture between the high
temperature server outlet air and the low temperature supply air.
They showthe total quantity of heat increase as air passes through
the IT servers inside a data center, and the heat quantity is calcu-
lated based on the volume and temperature deviation of air being
supplied to the IT server and air with increased temperature after
the heat removal frompassing throughthe servers. However, while
these indices can evaluate overall cooling efciency, it is difcult to
use themin making detailed cooling efciency judgments regard-
ing air re-circulation and by-pass prevention. Thus RCI and RTI,
which can most generally evaluate cooling performance and ef-
ciency based on temperatures at each server inlet and outlet, were
used for the analysis.
As mentioned before, SHI and RHI are macro indices that eval-
uate the overall efciency of IT server room by assessing the
efciency throughthe relationshipof the total heat suppliedindoor
and the total heat generated in removing the server rack heat. In
contrast, RCI is a micro index that evaluates thermal performance
by using the temperatures of the numerous server rack air inlets
and outlets. Therefore with this index, accurate judgment can be
made on the occurrence rate of air re-circulation and air bypass
near IT servers. However, since such evaluation is based only on
temperature conditions and evaluating the heat ow and thermal
equilibriumin the entire server roomis difcult, this index is inad-
equate for evaluating performance in terms of energy use. In other
words, it is outstanding in evaluating the temperature conditions
for IT server safety but somewhat decient in providing evaluation
for an ADS plan to save energy. Although RTI is clearly a powerful
index for checking air re-circulation and air bypass, since it calcu-
lates the total airow rate between the IT equipment and CRACs
without using numerical values presented by a guideline or stan-
dard, it has a weakness in that results which are relative depending
on the type of ADS can be produced.
4.1.1. Rack Cooling Index (RCI)
For a complete description of the RCI, the original work was
carried out by Herrlin [11]. This index deals with rack intake
temperatures; the conditions that air-cooled IT server depend on
for its continuous operation. The allowable and recommended
intake temperature limits in Fig. 9 represent the design condi-
tions, whereas the recommended limits refer to the preferred
facility operation. Over-temperature conditions exist once one or
more intake temperatures exceed the maximum recommended
temperature. The total over-temperature represents a summation
of over-temperatures across all rack inlets. Similarly, under-
temperature conditions exist when intake temperatures drop
belowthe minimumrecommended. The numerical values of these
limits depend on the applied guideline [8,9,20]. Table 3 provides
environmental conditions for electronic equipment and for facility
operation. Four environmental classes are described, together with
the environmental specications. The table is delineated for both
product operation and product power off. The product operating
environmental conditions, including allowable and recommended
values, refer to the state of the air entering the electronic equip-
ment. The conditions for classes 1 through 4 are the result of
consensus among the many environmental specications of man-
ufacturers of IT equipment.
Meaningful measure of rack cooling that alsocanbe represented
graphically. Focus onover-temperature, the temperature above the
maximumrecommended temperature. The indices should have an
easily understood scale from 0% to 100%, where 100% means that
all racks are cooled per some guideline or standard. Indication of
potentially harmful thermal conditions should be provided in a
simple but meaningful manner. The indices should be independent
of unit of measurement. Also, they should work with any guide-
line or standard that species maximum/minimumrecommended
and maximum/minimumallowable temperatures. In this way, the
Table 2
Simulation models physical specications and boundary conditions.
Size Quantity Air volume Remark
Supply perforated tiles 0.4m0.4m 80 tiles 570 CMH/tile
Return grilles 0.5m0.5m 20 grilles 2280 CMH/grille
CRAC units 2 EA 22,800 CMH/CRAC 16

C (supply)
IT equipments (4 shelves) 0.8m(W) 1.0m(D) 2.1m(H) 60 EA 650 CMH/rack
Aisle partition walls Height: 0.6m 60 EA
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
2152 J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155
Fig. 8. Temperature distribution for selected cooling systems at 4kW/rack. (a) SystemA: typical under-oor air cooling systemand (b) systemB: under-oor air cooling with
aisle partition system.
indices become a relative measure to the guideline or standard
used.
The RCI consists of twoparts, describingthe ITsever roomhealth
at the highendandat the lowendof the temperature range, respec-
tively. The RCI
HI
denition is Eq. (6) [11]:
RCI
HI
=
_
1
Total over temperature
Max allowable over temperature
_
100 [%]
=
_
1

(T
i
T
Max-rec
)
T
i
>T
Max-rec
n (T
Max-all
T
Max-rec
)
_
100 [%] (6)
The RCI
HI
is a measure of the absence of over-temperatures;
100% means that no over-temperatures exist. The lower the per-
centage, the greater probability (risk) that equipment experiences
temperatures above the maximum allowable temperature. The
interpretation of the index is as shown in Table 4. An analogous
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155 2153
Table 3
Equipment environment specications [9].
Class Product operation
a, b
Product power off
b, c
Dry-bulb temperature (

C) Relative humidity (%) Dry-bulb temperature (

C) Relative humidity (%)

Allowable Recommended Allowable Recommended
1 1532
d
2025 2080 4055 545 880
2 1535
d
2025 2080 4055 545 880
3 532
d, e
NA 880 NA 545 880
4 540
d, e
NA 880 NA 545 880
a
Product equipment is powered on.
b
Tape products require a stable and more restricted environment (similar to class 1). Typical requirements: minimum temperature is 15

C, maximum temperature is
32

C, minimumrelative humidity is 20%, maximumrelative humidity is 80%, maximumdew point is 22

C, rate of change of temperature is less than 2

C/h, rate of change

of humidity is less than 5% RH per hour, and no condensation.
c
Product equipment is removed fromthe original shipping container and installed but not in use, e.g., during repair, maintenance, or upgrade.
d
Derate maximumdry-bulb temperature 1

C/300mabove 900m.
e
With a diskette in the drive, the minimumtemperature is 10

C.
index can be dened for temperature conditions at the low end of
the temperature range. The RCI
LO
is a complement tothe previously
dened index especially when the supply condition is below the
minimum recommended temperature. Under such circumstances
the two indices can preferably be used in tandem. On the other
hand, if an under temperature is of less concern, the focus should
be on maximizing the RCI
HI
. The RCI
LO
is dened as Eq. (7) [11]:
RCI
LO
=
_
1
Total under temperature
Max allowable under temperature
_
100 [%]
=
_
1

(T
Min-rec
T
i
)
T
i
<T
Min-rec
n (T
Min-rec
T
Min-all
)
_
100 [%] (7)
4.1.2. Return Temperature Index (RTI)
In the data centers of the open architecture, the RCI
HI
can be
improved by increasing the supply airow rate and lowering the
supply air temperature. However, these measures are associated
with an energy penalty. The RTI is a measure of the energy perfor-
mance of the air management system. The index is dened as Eq.
(8) [14]:
RTI =
_
T
return
T
Supply
T
Equipment
_
100 [%] (8)
TheRTI is alsotheratioof total airowthroughtheIT-equipment
to the total airowthroughthe CRACs. Avalue above 100%suggests
net recirculation air, which elevates the return air temperature.
Unfortunately, this also means elevated equipment intake tem-
peratures. A value below 100% suggests net by-pass air; cold air
by-passes the electronic equipment and is returned directly to the
CRAC, reducing the return temperature. This may happen when the
supply airowis increased to combat hotspots or if there are leaks
in the raised oor. There might be a number of legitimate reasons
to operate belowor above 100%. For example, some air-distribution
schemes are designed to provide a certain level of air mixing (recir-
Fig. 9. Denition of total over temperature and total under temperature.
culation) to provide an even equipment intake temperature. The
interpretation of the index is shown in Table 5.
4.2. Application of the RCI and RTI
4.2.1. IT server of intake temperatures
Although CFD simulation allows visualization of temperatures
across the IT server room, organizing the cooling effectiveness of
different systemoptions can be challenging. By using the CFDtech-
nology, the rack intake temperatures can accurately be established
by using monitor points much the same as the method of taking
eld measurements. To evaluate the cooling efciency of the aisle
partition system, a comparison was made with the under-oor air
cooling system by using the RCI based on the temperature distri-
butions deduced from the CFD simulations (Section 3). For this, a
total of 240 intake temperatures were taken for 60 rack servers.
Fig. 10 shows the resulting rack intake temperature plots for the
two ADSs at the heat densities: 4kW/rack. The intake temperatures
are arranged in order of increasing temperature and the intakes are
numbered accordingly. In addition, the temperature steps corre-
spond to the upper equipment shelf levels. Obviously, the matched
airowdoes not manage to submerge the top shelves with the cold
supply air due to loss of chilled air at the end of the cold aisles.
For the aisle partition system, the temperature distribution is rel-
atively at; the temperature variation from equipment intake to
intake is small (XZ section of Fig. 8 also suggests that the aisle
temperature is even). Indeed, this gravity assisted mixing may be
the most important difference between the typical under-oor air
cooling system and the aisle partition system. Note that although
the supply temperature is higher than for the typical under-oor
air cooling system, the peak temperature is lower. These temper-
ature distributions provide a complete picture of the thermal rack
environment. However, a well-selected index would help compare
the systems based on the very same criteria.
4.2.2. Return air temperatures to CRAC units
Because the temperature rise across the IT server provides the
potential for high return temperatures, it makes sense to normalize
the RTI with regard to this entity. In other words, the RTI provides
a measure of the actual utilization of the available temperature
Table 4
Rating of Rack Cooling Index (RCI).
Rating RCIHI
Ideal 100%
Good 100%
Acceptable 9195%
Poor 90%
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
2154 J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155
Fig. 10. Rack intake temperature for typical under-oor air cooling and the aisle partition system.
Table 5
Rating of Return Temperature Index (RTI).
Rating RTI
Target 100%
Re-circulation >100%
By-Pass <100%
differential. Consequently, a lowreturn air temperature is not nec-
essarily a signof poor air management. If the ITserver only provides
a modest temperature rise, the return air temperature cannot be
expected to be high.
RTI was calculated by using the simulation result values of the
return air temperature emitted after eliminating the rack server
heat at each return air grille position shown in Fig. 6. Table 6 shows
the return air temperature, supply air temperature and tempera-
ture rise across the IT equipment used for calculating RTI.
In general, many IT equipments or rack servers have a tem-
perature rise of about 15

C, whereas new blade servers can have

a temperature differential of 40

C. This study focused on high-

density data centers that use normal IT servers, and as mentioned
earlier, the analysis model was based on the 4kW/rack heat den-
sity standard. Although lately high performance blade servers are
being developed, so far installation of such servers is limited to a
fewdata centers. However, since there is a high possibility of blade
server application becoming ubiquitous in the future, such addi-
tional research on 10kW/rack and greater is needed. But with the
current general air distribution system, it is realistically impossible
to provide thermal management of data centers operating blade
servers.
4.3. Evaluation of air distribution systems thermal performance
and cooling efciency
Based on CFD modeling, Fig. 11 shows the results as measured
by the RCI and RTI. These index results, calculated by using the IT
servers air intake temperature and the average return air temper-
Fig. 11. The Rack Cooling Indices (RCIs) and Return Temperature Indices (RTIs) for
selected cooling systems with environmental class 1 as the guideline.
ature at return air grilles of the IT server room, are given. Compared
to the typical under-oor air cooling system, the aisle partition sys-
temimproves the thermal conditions, that is, higher RCI
HI
. The aisle
partitioned design provides a striking boost to the cooling perfor-
mance of the raised oor. At the 4kW per rack level, the RCI
HI
is
elevated to 99% (ideal). Since the RTI is above 100%, some air re-
circulates the IT equipment. This demonstrates that an RCI analysis
should be accompanied by an energy analysis; in this case by using
the RTI. As shown in the resulting graph, when the aisle partition
systemis installed, the air re-circulation can be prevented to some
extent in comparison with the existing system. That is, when the
aisle partition system is installed, although thermal performance
improves by about 10% for RCI
HI
and 8% for RTI, there is still a
slight amount of air re-circulation occurring because of the open
traditional architecture. However, since the aisle partition system
provides larger energy savings for RCI
HI
, it can be controlled with
Table 6
Return air temperature for selected cooling systems.
TReturn (

C) SystemA: typical under-oor air cooling system SystemB: under-oor air cooling with aisle partition system
RA-1 RA-2 RA-3 RA-1 RA-2 RA-3
Y=2.5m 28.1 28.3 27.9 30.8 31.0 30.9
Y =8.0m 27.9 28.8 28.9 30.9 29.0 30.8
Y =13.5m 28.2 27.2 28.5 30.6 30.2 29.1
Cooling systems conditions: RA-#: return air grille (Fig. 6). T
Supply
=16

C. T
Equip
=15

C.
Journal Identication = ENB Article Identication = 3205 Date: July 19, 2011 Time: 7:36pm
J. Cho, B.S. Kim/ Energy and Buildings 43 (2011) 21452155 2155
the supply air from the CRAC unit at a higher temperature and
is deemed to be superior in terms of applicability. That is, it can
increase the chillers COP by raising the supply chilled water tem-
perature and thereby serve as a measure for reducing the input
energy consumed. The RCI and RTI metrics provide yardsticks for
ranking design permutations. Each engineer needs to determine
the proper balance between the rack cooling effectiveness, energy
performance, and other considerations. Although RCI and RTI are
indices that can objectively evaluate thermal performance, they
are completely different fromeach other. RCI focuses on IT equip-
ment protection, whereas RCI concerns thermal management for
minimizing the indoor air re-circulation and air bypass. Since both
elements areimportant, it is important todetermineanappropriate
balance between the two. Because the ultimate goal of data center
thermal management is the protection of IT equipment, it is desir-
able to rst satisfy the minimum conditions in terms of RCI while
improving RTI that can increase energy efciency.
5. Conclusion
This paper describes two performance metrics for analyzing air
distribution systems in high density data centers. Given that air
distribution from a raised oor is the most common way of cool-
ing data centers, this scheme was used to demonstrate the use and
benets of the metrics. The Rack Cooling Index (RCI) and Return
Temperature Index (RTI) were computed with CFD simulation to
evaluate the thermal equipment environment in two cooling sys-
tems. For typical under-oor air cooling systems, withgood cooling
efciency, it was observed that air re-circulation takes place in the
upper part of the server (2.12.7m). To prevent this, the study pro-
posed the aisle partition system that physically divides the cold
aisle and the hot aisle to the 0.6m level of the upper part of the
server. By utilizing the RCI, it was shown howthe cooling efciency
of the under-oor air cooling systemcanbe signicantly boostedby
partitioning the cold aisles. By the aisle partition walls, however,
the airow distribution becomes less important and the airow
can be increased to levels that would have been prohibitively high
in conventional under-oor air cooling systems. The CFD simula-
tion results show that the high and low temperature difference
by IT server height is less than 10

C when aisle partition walls

were installed, but when they were not installed, the difference
can increase to a maximum of 15

C, while the return air tem-

perature is reduced by about 1.4

C from the prevention of air

re-circulation. Normally air by-pass and re-circulation account for
a 30% overall loss of efciency in a data center. We therefore begin
with the idea that we want to separate cold and warm air opti-
mally. Lastly, the analysis basedonthe coolingefciencyevaluation
indices showed a performance improvement of about 10% for RCI
HI
and 8% for RTI. Because the aisle partition systemcan also raise the
supply air temperature, it is possible to improve the energy perfor-
mance connected to the primary system. This study is expected
to have value in providing more objective data for data center
managers or engineers in judging the applicability of an aisle par-
tition system and at the same time presenting an alternative for
efcient HVAC systemoperation. To optimize the design of the sys-
tem, further study is needed in the future that takes into account
the height and the form variation of the aisle partition wall as
variables.
References
[1] C. Kurkjian, J. Glass, Meeting the needs of 24/7 data centers, ASHRAE Journal
49 (2) (2007) 2435.
[2] H.S. Sun, S.E. Lee, Case study of data centers energy performance, Energy and
Buildings 38 (5) (2006) 522533.
[3] K.C. Karki, S.V. Patankar, Airowdistribution through perforated tiles in raised-
oor data centers, Building and Environment 41 (2006) 734744.
[4] A. Beitelmal, C. Patel, Thermo-uids provisioning of a high performance high
density data center, Distributed and Parallel Databases: High Density Data
Centers 21 (2/3) (2007) 227238.
[5] C.E. Bash, C.D. Patel, R.K. Sharma, Efcient thermal management of data centers.
Immediate and long-term research needs, International Journal for HVAC&R
Research 9 (2) (2003) 137152.
[6] C. Patel, R. Sharma, C. Bash, A. Beitelmal, Thermal considerations in cooling
large scale high computer density data centers, in: Proceedings of ITHERM
2002 the Eighth Intersociety Conference on Thermal and Thermomechanical
Phenomena in Electronic Systems, 2002, pp. 767776.
[7] M.K. Herrlin, A new tool for evaluating and designing the thermal environ-
ment in telecomcentral ofces, in: Proceedings of Telecommunications Energy
Conference, 2006, pp. 15.
[8] Telcordia (M.K. Herrlin), Generic Requirements NEBS GR-3028-CORE, Thermal
Management in Telecommunications Central Ofces, Telcordia Technologies,
2001.
[9] TC9.9, MissionCritical Facilities, TechnologySpaces, andElectronic Equipment,
Thermal Guidelines for Data Processing Environments, American Society of
Heating, Refrigerating and Air-Conditioning Engineers Inc., 2004.
[10] R.K. Sharma, C.E. Bash, C.D. Patel, Dimensionless parameters for evaluation
of thermal design and performance of large scale data centers, in: Proceed-
ings of AIAA2002-3091, American Institute of Aeronautics and Astronautics
Conference, 2002.
[11] M.K. Herrlin, Rack cooling effectiveness in data centers and telecom central
ofces: the rack cooling index (RCI), ASHRAE Transactions 111 (2) (2005) 111.
[12] J.W. VanGilder, S.K. Shrivastava, Capture index: an airow-based rack cooling
performance metric, ASHRAE Transactions 113 (1) (2007) 126136.
[13] J. Cho, T. Lim, B.S. Kim, Measurements and predictions of the air distribution
systems in high compute density (Internet) data centers, Energy and Buildings
41 (10) (2009) 11071115.
[14] M.K. Herrlin, Improved data center energy efciency and thermal performance
by advanced airowanalysis, in: Proceedings of Digital Power Forum, 2007, pp.
1012.
[15] M.K. Herrlin, Airow and cooling performance of data centers: two perfor-
mance metrics, ASHRAE Transactions 114 (2) (2008).
[16] U.S. Department of Energy, U.S. Data Centers Save Energy Now, Industrial Tech-
nologies Program, DOE-Energy Efciency and Renewable Energy, 2009.
[17] J. Cho, T. Lim, B.S. Kim, Cooling systems for IT environment heat removal in
(Internet) data centers, Journal of Asian Architecture and Building Engineering
7 (2) (2008) 387394.
[18] EmersonNetworkPower, Energylogic: ReducingDataCenter EnergyConsump-
tion by Creating Savings that Cascade Across Systems, A White Paper fromthe
Experts in Business-Critical Continuity, 2008.
[19] TC9.9, MissionCritical Facilities, TechnologySpaces, andElectronic Equipment,
Design Consideration for Datacom Equipment Centers, American Society of
Heating, Refrigerating and Air-Conditioning Engineers Inc., 2005.
[20] Telcordia, Generic Requirements NEBS GR-63-CORE, NEBS Requirements:
Physical Protection, Telcordia Technologies, 2002.