No detectable additional cost was associated with switching on CICS® IA data collection, with all collection options set to N.
Table 1 shows the CPU cost in microseconds of collecting interdependency data for a selection of CICS commands, using the configuration described in How IA affects performance. Use the Large Systems Performance Reference ratios to make the necessary adjustments for your machine environment.
API command | Category | CPU cost with category option set to Y | CPU cost with category option set to D |
---|---|---|---|
READ | FILES | 7.0 | 8.0 |
READQ TS | TS QUEUES | 6.4 | 7.6 |
READQ TD | TD QUEUES | 6.1 | 7.1 |
LINK | PROGRAMS | 10.2 | 18.1 |
For the PROGRAMS category, CICS IA exits are called on the LINK and also on the RETURN from the LINK. Hence the CPU cost is higher than for other categories. Although the CICS IA costs are not identical for each CICS command, with the exception of the LINK command, they are similar.
API request | Category | CPU cost with category option set to Y |
---|---|---|
SELECT | DB2 | 12 |
GET and PUT | MQ | 10 |
DB2 and MQ requests run on L8 TCBs. In multiprocessor environments, work can be run concurrently on the QR and multi-L8 TCBs. With some workloads, you might see a detectable increase in hardware instruction and data cache misses. The rate of cache misses affects the CPU costs. Thus, the figures shown in the table above might vary, depending on your workload and the machine environment.
However, some options enable commonly used exits if selected. For example, if the Programs option is set to Y for interdependency data collection, XEIIN and XEIOUT exits are enabled, and are driven for every CICS command encountered. Therefore, a performance overhead is associated with driving these exits, although no CICS IA data is collected by these exits for CICS commands that are not related to programs.
See CICS IA exits for further details.