Measuring zSeries System Performance Dr. Chu J. Jong School of Information Technology Illinois State University 06/11/2012 Sponsored in part by Deer & Company
Dec 24, 2015
Measuring zSeries System Performance
Dr. Chu J. JongSchool of Information Technology
Illinois State University06/11/2012
Sponsored in part by Deer & Company
Outline
• Computer System Performance• Performance Factors and Measurements• zSeries Performance– Measuring Application Performance– Measuring System Performance
• Additional Tools used• Discussion
Computer System Performance
• Amount of time used to complete a task• Amount of work completed in unit of time• Resource required and resource usage• Others– Storage– Channel– Scalability– Availability (MTBF and MTTF)– Power– Etc.
Factors of System Performance
Hardware• Processor• Bus• Clock• Memory• Secondary Storage• I/O Devices• Network• Cooling• Power
Software• Resource Allocation• Resource Sharing• Process Distribution• Operating Systems• Algorithms• Context Switches• Compilation• Optimization• Others … and more
Performance Measurement
• Metrics include: availability, response time, channel capacity, latency, completion time, service time, bandwidth, throughput, scalability, performance per watt, compression ratio, speed up, …, and more.
• Two are used:– Response Time– Throughput
Measuring zSystem Performance
• Application – DB2Compare the performance of accessing data stored in DB2 table against reading the same data accessed directly in VSAM running on z/OS hosted by IBM zSystem.
• System – CP, the HypervisorCompare and analysis the performance of resource management by Hypervisor against the performance of resource management by z/VM and Linux (guest) of z/VM hosted by IBM zSystem
Application – DB2
A Normalized Comparison of DB2 and Direct Access Performance under z/OS Environment(By Christopher Corso)
Compares the performance of accessing data stored in DB2 tables against reading the same data values accessed directly in VSAM files. Validation testing is performed on MVS mainframes
running DB2 version 9 under zOS. The comparison of the performance will be of a DB2 FETCH with a read until end of file on a direct access VSAM file. The resulting CPU processing times
of the different methods are discussed and conclusions are offered
Testing and System Configuration
Time differentials of:• Task Control Blocks (TCB)• Service Request Blocks (SRB) • Computer clock speeds (CPU)Systems Used• ISU Mainframe (z890) – zOS, DB2 version 9• IIC Mainframe (IBM Innovation Center-Dallas) VM– zOS, DB2 version 9
Testing Programs and Names
• DB2 only• VSAM only• DB2 with internal files• VSAM with internal files• SPEED1 – DB2 processing only• SPEED2 - direct access VSAM processing only• SPEED3 - direct access VSAM with internal files• SPEED4 - DB2 with internal files.
Testing Results
• Wall Clock Time• Task Control Block (TCB) Time• Service Request Block (SRB) Time• CPU Time
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 450
10
20
30
40
50
60
Wall Clock Time
ISU SPEEDA1ISU SPEEDA2ISU SPEEDA3ISU SPEEDA4
Run number
Tim
e in
seco
nds
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 270.000000
2.000000
4.000000
6.000000
8.000000
10.000000
12.000000
14.000000
16.000000
Wall Clock Time
IIC SPEEDA1IIC SPEEDA2IIC SPEEDA3IIC SPEEDA4
Run number
Tim
e (in
seco
nds)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 450
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
TCB time by run
ISU TCB SPEEDA1ISU TCB SPEEDA2ISU TCB SPEEDA3ISU TCB SPEEDA4
Run number
Tim
e in
Sec
onds
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 450
0.05
0.1
0.15
0.2
0.25
SRB time per run
ISU SRB SPEEDA1ISU SRB SPEEDA2ISU SRB SPEEDA3ISU SRB SPEEDA4
Run number
Tim
e in
Sec
onds
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 450
0.5
1
1.5
2
2.5
CPU time per run
ISU CLOCK SPEEDA1ISU CLOCK SPEEDA2ISU CLOCK SPEEDA3ISU CLOCK SPEEDA4
Run number
Tim
e in
Sec
onds
Observation
• DB2 proved faster than direct VSAM access• One method is not significantly faster to process
than the other• The least practical method of storing the data is
to put it in variables within the source code itself• Converting direct access VSAM is not always the
best option• It is much easier for users of the data to access it
via DB2 tables
System – CP Hypervisor
compare and analysis the performance of resource management by Hypervisor against the performance of resource management by z/VM and Linux (guest) of z/VM hosted by IBM zSystem. The
purpose is to analyze and correlate the relationship between the resource management of Guest Virtual Machine (Linux on z/VM)
and the hypervisor of hosting Virtual Machine (z/VM). We will run benchmark on combinations of different of processes and guest VMs to collect their performance data for a closure statement.
Optimizing Guest System’s Addressing Space in a Virtual Environment
By Niranjan Sharma
Performance of CP and z/VM
• Resource Allocation– By CP, By z/VM, By guest O/S – Linux
• Memory Management– By CP, By z/VM, By guest O/S – Linux
• CPU Cycle Distribution– By CP, By z/VM, by guest O/S – Linux
• Mainframe Resource Utilization and Scalability– Do they fit in the distributed system?– How about Cloud Computing and Parallel Computing
Performance Concerns• LPAR Optimization
– How many LPARs becomes too many– What are the overheads of managing LPARs
• Guest Optimization– How many guest O/S’s becomes too many– What are the overheads of managing guest O/S’s
• Process Optimization– How many processes a guest O/S may handle to maintain scalability– What are the cost of context switches
• Resource Sharing– Processor assignment– Memory allocation– Buffer space and channel distribution
Benchmark Testing
• CPU Intensive Application Response Time and Throughput– Scaling from 1 to 2n processes per guest O/S– Scaling from 1 to 2m guests per LPAR– Scaling from 1 to 2k LPAR per system
• Memory Intensive Application Response Time and Throughput– Scaling from 1 to 2n processes per guest O/S– Scaling from 1 to 2m guests per LPAR– Scaling from 1 to 2k LPAR per system
• Resource Utilization and Scalability• LINPACK test suite – parallel computing
Discussions
• Under what circumstance CP allocate its resource adequately
• Under what circumstance VM manage its resource effectively
• Scalability of CPU intensive applications• Scalability of memory intensive applications• Mainframes v.s. Distributed Systems – A
Collaboration Approach