Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
L3C 1
L3C 0
MCU
L3 Data L3 DataGX
Toni Sacconaghi of Bernstein Research estimates that 40% of IBMs profits are mainframe-related.
Little to no reference to
Multi-core/SMT
Dynamic compilation/managed runtimes
I/O attached accelerators
domain-specific devices/languages
distributed computing
parallel programming models
etc
XCAT
Scripting Ecosystem
WebSphere
CICS/IMS
DB2
Scoring Rules
SPSS ILOG
WebSphere
SPSS CICS/IMS
CPLEX DB2
InfoSphere
Infoserver
Cognos
InfoSphere
Warehouse
Scoring Rules
SPSS ILOG
WebSphere
SPSS CICS/IMS
CPLEX DB2
InfoSphere
Infoserver
Cognos
InfoSphere
Warehouse
=> ~37%
Base: North American and European enterprise software developers; Source: Forrsights Developer Survey, Q1 2013
=> ~37%
=> ~26%
Base: North American and European enterprise software developers; Source: Forrsights Developer Survey, Q1 2013
Scoring Rules
SPSS ILOG
WebSphere
SPSS CICS/IMS
CPLEX DB2
InfoSphere
Scripting Ecosystem
Infoserver
Cognos
InfoSphere
Warehouse
Batch windows
Fixed elapsed window to complete
eg. Balance books overnight before opening for business
next day
A range of aggressive hardware and systems designs IBM DB2 Analytics Accelerator
Fit-for-purpose/Hybrid systems
Appliances (eg. Netezza)
GPUs/FGPAs
Co-processors (eg. crypto)
Transactional memory
Middleware/Application Middleware/Application
10.00%
0.00%
Latency CPU Consumption per tran TCP TCP
-5.00%
IP SMC-R IP SMC-R
-10.00%
P ercent Change
Interface Interface
-30.00% IP Network
TCP connection establishment over IP
(Ethernet)
Dynamic negotiation for SMC -R
-35.00%
-40.00%
-40.00%
-45.00%
48x faster !
2x faster
Today:
Optimize once for single static architecture
Synergy with micro-architecture is open-loop
Parallelism is in its infancy
Some dynamic in Java (Just-in-time)
z/OS Multi-Threaded 64 bit Java Workload 16-Way
~12x Improvement in Hardware and Software
160
zEC12 SDK 7 SR3
Tomorrow: 140
Aggressive +
LP Code Cache
zEC12 SDK 7 SR1
Normalized Throughput
120
Deeper synergy with micro-architecture 100 z196 SDK 7 SR1
RIOFF RION
// C code
Method B
J9 Runtime Call void processBuffer() {
(e.g., interpreter)
}
Execute Update
Sampling thread
21 2013 IBM Corporation
RI Example: Path Splitting/Specialization
B1 B1
compiler
B2 B3 B2 B3
x=i x=4 x=i
x=4
B4 B4
B4
B5
j = y >> 2
B5 B6
j = y/x B7 B5
j = y/x
B7
Path can be specialized
22 2013 IBM Corporation
Binary Optimization
Technology that enables re-optimization of legacy COBOL binaries
on the latest System z without requiring source-level re-compilation.
legacy COBOL
binary with ESA390
Binary Optimization technology being built by IBM Research (Tokyo) instructions
Solution for the Dusty Deck Problem: Can re-optimize the existing
large body of legacy code that the customers are unable and/or
unwilling to recompile
Built on top of IBM Testarossa Optimizer
Value
Binary
Upgrade the binary from really old ISA to the exact ISA of the machine
Optimizer
the code is running on
Inject the latest COBOL optimization technology into legacy programs
Ability to start utilizing profiling and RI-based feedback into to optimize
the program
Experimental Results* optimized COBOL
up-to 4.62x and average 1.89x over the original binary on z196 binary with zEC12
up-to 3.31x and average 1.94x over the original binary on zEC12 instructions
Alpha-level prototype available on developerWorks
* Measured using twelve benchmarks in the internal testsuite, used by the COBOL compiler dev. team.
233 2013 IBM Corporation
2013 IBM Corporation
Concurrency and Parallelism in the Enterprise
node
node
tail
En-queue
last
node
read_hash(key) { read_hash(key)
Wait_for_lock(); TRANSACTION_BEGIN
T
T
Thr2: read_hash()
T h ro u g h p u t (o p s/sec)
Thr3:read_hash()
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Threads
http://www.dzone.com/links/presentation_languagelibraryvm_coevolution_in_jav.html
Scoring Rules
SPSS ILOG
WebSphere
SPSS CICS/IMS
CPLEX DB2
InfoSphere
Scripting Ecosystem
Infoserver
Cognos
InfoSphere
Warehouse
Creating an Open Community to drive full stack innovation for the cloud: OpenPOWER
Systems
Industrys first open system design for cloud data
centers Processors
Semiconductor
Custom development group for hyperscale Technology
A A
GC
D D
C
hot
B C B
In cache/tlb
Java objects
Out of cache/tlb
32 2013 IBM Corporation