View Only

MQ for z/OS on IBM z16 - What performance benefits might you see?

By Anthony Sharkey posted 8 days ago


IBM announced the new IBM z16 mainframe server in 2022, and what follows is an overview of what MQ for z/OS' performance expectations were, plus the actual results of moving our MQ performance sysplex from z15 to z16.

MQ performance reports are generally posted on the mqperf github repository and the MQ for z/OS on IBM z16 performance report is no different, and is available here.

When setting expectations of any performance improvements on z/OS, the first place to look is typically the IBM Large System Performance Reference (LSPR) website. Consideration should be given to the complexity of the workloads being run - and the measurements that we run vary significantly from very simple to complex, so there wasn't a single percentage value that we could apply to our workloads. Despite this, we used a rule of thumb of 9 to 14% reduction in transaction cost - although of course, some measurements could be more and some could be less.

Processor changes

On IBM z15, each drawer has 4 single chip modules (SCM), and each SCM has has 12 cores, of which 9 to 11 will be available. On our performance system, we were allocating up to 10 cores per processor unit.

On IBM z16, each drawer has 4 dual chip modules (DCM), where each DCM has 9 to 11, or 10 to 15 cores. On our performance system, we were allocating up to 6 cores per (single) chip.

How do the processor changes affect MQ micro-benchmarks

Across the MQ performance micro-benchmarks, over one third of the measurements achieved or exceeded the improvements expected by the LSPR, with three quarters of the measurements demonstrating transaction cost lower than on IBM z15.

Of the remaining measurements, where the transaction cost was higher on IBM z16 than IBM z15, the measurements fell into 2 categories:
  1. Very simple tests with short pathlengths and small messages (2KB or less).
  2. Tests where on IBM z15 the workload could be contained on 1 processor unit, whereas on IBM z16 may span 2 processor units. For example, a large proportion of the tests with increased cost were run on LPARs with 10 CPUs allocated, so on z15 this would be on a single SCM, whereas on z16 this would be on a single DCM but would require both chip modules.

CFCC level 25

The introduction of CFCC level 25 requires an additional 8MB of storage in both the CSQ_ADMIN and APPLICATION structures to achieve the same capacity as earlier CFCC levels.

Of more note is that the CSQSYSAPPL structure which may need to grow beyond the INITSIZE of 20,000 as suggested in SCSQPROC(CSQ4CFRM).

A snapshot of MQ performance benchmarks on IBM z16

Whilst the report "MQ for z/OS on z16" gives more details on how MQ for z/OS performs on the newest IBM mainframe, it may be of interest to show some of the highlights:

  • Scalability of non-persistent in-syncpoint workloads achieving rates of up to 11% with cost reductions of up to 13%.
  • Channels protected with TLS ciphers achieving cost reductions of up to 14%.
  • Channels using hardware compression via COMPMSG(ZLIBFAST) achieved cost reductions of up to 14% over IBM z15.
  • Queues protected with AMS policies achieving transaction cost reductions of up to 18%.
  • Persistent message measurements demonstrate cost reductions of up to 28% over IBM z15.