Task collection frequency: Performance results
In CICS Interdependency Analyzer (IA) V5.2, we introduced the capability of collecting dependency data for every nth task. For more information about this feature, see 2.1.1, “Using the Optimum Collection option” on page 24.
When developing this feature, we asked our CICS performance team in Hursley to perform some measurements against the new feature. This appendix contains a brief description of the tests and the results.
Test environment
The tests were performed with an internal workload, which consisted of the following specifications:
• A COBOL BMS application with VSAM IO
•Running in two TORs routing to two AORs routing to an FOR
•With a constant transaction rate of 3800 transactions per second
•On zEC12 HA1 system, which is equivalent to a 2827-716
The following software was used:
•z/OS V2.1
•CICS TS V5.2
•CICS IA V.2
The following CICS IA options were used:
•The CICS IA collection files were shared by using VSAM RLS
•The collector was set up to collect interdependency data
•The “usage count” option was switched on
•Dynamic calls were monitored
•All CICS APIs and SPIs were monitored
The following test measurements were used:
•Used IBM RMF™ data to measure the overall CPU and transaction rate at 5-minute intervals
•The CICS IA collection frequency was varied by using the “Trigger for Task collection” option
Test results
The chart in Figure B-1 on page 237 shows the processor usage of running the CICS IA collector. The column on the left shows the CPU with CICS IA switched off. The column on the right shows the CPU processor usage for when collecting dependency data for every task. The columns in the middle show the processor usage when collecting data for every “n” task where n is 5, 10, 50, 500, and 9999. You can see that when we set the “Trigger for Task collection” value to 50 or greater, there is no great difference in the processor usage.
|
Note: These test results are based on an application workload that consists of EXEC CICS API and SPI calls with a minimum of business logic. This processor usage is not typical of a customer environment where an API call would be followed by business logic.
|
Figure B-1 CICS IA processor usage for the sample workload
The results are also shown in Table B-1.
Conclusions
The following conclusions were reached:
•Enabling CICS IA dependency collector always has an associated cost of driving the user exit, regardless of the collection frequency.
•For this sample workload, reducing the task collection frequency to lower than 1 in 50 has little or no effect.
•Collecting data for one in every 10 tasks can reduce the CICS IA processor usage by 65%.
The processor usage for the sample workload is shown in Table B-1.
Table B-1 CICS IA processor usage for the sample workload
|
Collect every ‘n’th task
|
Base workload
(CPU ms)
|
CICS IA
processor usage
(CPU ms)
|
Savings in CPU
|
|
IA Off
|
0.225
|
0.000
|
-
|
|
9999
|
0.225
|
0.027
|
76%
|
|
500
|
0.225
|
0.026
|
76%
|
|
50
|
0.225
|
0.030
|
73%
|
|
10
|
0.225
|
0.038
|
65%
|
|
5
|
0.225
|
0.048
|
56%
|
|
1
|
0.225
|
0.110
|
-
|
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.