Testing Datacom SMPTASK_USING_IEAV MUF Startup Option

Products

• Datacom
• Datacom/DB
• Datacom/AD

Issue/Introduction

Planned implementing the MUF Startup Option SMPTASK_USING_IEAV. How can the effect be tested on the billable CPU before and after the change?

Resolution

The information regarding the use of the TCBs and SRBs is reported during the MUF shutdown, and on demand with the ALL_INFO_REPORT console command. Since changes to turn this feature on or off can only be made in the MUF Startup Options, the MUF will need to be recycled, and so the MUF shutdown report is the best way to determine the effectiveness. If the MUF stays running for a long time (some run for months before shutting down), the statistics mentioned below can be seen by issuing the console command /F mufname,ALL_INFO_REPORT pxxdd (where pxxdd is one of the SYSOUT-defined DD statements: PXXDD, PXXCBS, PXXSQL, or PXXDST for use with the MUF Startup Option called SYSOUT, or you can use the PXX file if not using the SYSOUT MUF Startup Option).

There is a section of the shutdown report titled "TCB/SRB USE SUMMARY INFORMATION" that contains the metrics about zIIP and CPU usage. In addition to the statistics about each TCB and SRB, there is a total line for this section, where you can see the effect of this change on the billable CPU. Below is a snapshot of a workbook showing the "Before" values and another showing the "After" values, with an analysis following the pictures.

To see the effect of the change to the billable CPU amount from using SMPTASK_USING_IEAV YES, it is best to compare equivalent workloads. However, since usage is not predictable nor manageable, you can observe the trend (note that in our example, workloads were designed to be very similar). For the billable CPU amount, we need to look at the beige/brown numbers.

The Total CPU before the change showed about 40 seconds, and after the change 30 seconds. This represents about a 25% reduction in overall CPU from this change.
For zIIP processing, before the change, zIIP processors accounted for about 78% of the total CPU (zIIP CPU/Total CPU), and after the change, an increase to 89% of the total, which is about a 14% increase in zIIP usage.

When looking at the workload contained within the MUF (excluding system-level processing billed to the MUF, like XCF management, memory paging, etc.), we need to look at the blue numbers. Here, even though the zIIP percentage is nearly the same (94% before vs 93% after), the total CPU dropped from 33 seconds to 29 seconds, about a 12% improvement.

To make this analysis easier, there is a copy of the Microsoft Excel workbook attached, showing this demo information, along with a separate worksheet for your use.

Additional Information

There are a number of things that can be done to improve processing efficiency of a Datacom MUF, depending on the objective to be reached. Some things, like the MUF's execution priority, WLM setting, and processor or I/O resource contention can be managed outside the MUF. Others, like region size, memory usage, or data compression and caching are managed within the MUF and its data configurations. This article looks specifically at using SMPTASK_USING_IEAV as one way to reduce billable CPU usage.

For more information about this, please refer to the section titled "Reducing Billable CPU for Every MUF in Your Environment" in the Datacom/DB 15.1 documentation.

As always, please contact Broadcom support for Datacom if you have further questions.