Whilst many of the performance improvements in the CICS TS V5.3 open beta have focused on specific ways into CICS, such as web services and JSON requests through z/OS Connect, there have been improvements in other areas that will give a benefit to many other types of workload.
MRO connections and high numbers of primary & secondary sessions
Some of those improvements have come about because for every CICS TS release the performance team do hot spot analysis on different areas of function. Hot spot analysis shows areas of code that are used a lot more than others. If the algorithms used at those hot spots have potential for improvement it can result in reduced CPU usage and better throughput. One such area that has been reviewed is various configurations of MRO connections between regions under workload. In one configuration, where a high number of primary and secondary sessions were defined, a hot spot was seen. After reviewing the algorithm used, that algorithm was made more efficient when searching for a suitable session. Our performance numbers showed that a configuration with the maximum number of primary and secondary sessions defined saw the largest CPU reduction.
New hardware, new possibilities
The new prerequisite of a z9 machine has enabled the CICS TS V5.3 open beta to exploit the hardware instructions it introduced including store clock fast (STCKF). Although only one instruction, in areas where a lot of timing is involved when combined with other performance tweaks these improvements all add up. Some of those other performance tweaks include the cache alignment of some key control blocks in the kernel, storage manager, monitoring and other areas, meaning CICS makes more efficient use of memory. Key beneficiaries of these enhancements are both monitoring and tracing in CICS where there is a measurable improvement, where less CPU is used by these features when they are enabled. Whilst the z9 is the prerequisite, our performance team have seen considerable reductions in CPU usage when running on the new z13. One example of this is comparing a web services workload doing full SSL handshakes on a zEC12 vs a z13 saw a 10% reduction in CPU usage, showing that staying current on hardware as well as software will give you the best benefit.
CPU monitoring made simpler
One key piece of information when gaining insight into the performance of a CICS region, and the work running within it, is how much CPU, both general purpose and zIIP, that the region has been using on average per transaction. In the past this required the CICS Monitoring Facility performance class to be turned on and information from the SMF 110 records to be extracted and totalled. In the open beta there is no longer a need to have CICS monitoring enabled, as the totals for GP and offloaded CPU usage for all completed transactions can be found in the monitoring domain global statistics. This means that this information can be obtained without requiring many SMF 110 records to be written, stored, and processed.