Workload Manager - velocity

vasanthz · Posted: Wed Jul 31, 2013 2:58 pm

Hello,
I am trying to learn some basics of mainframe performance,
There was a term "velocity" allocated to workloads,
The manual says,
Velocity: This is a measure of how fast work should run when ready, without being delayed for resources. Velocity is a percentage from 1 to 99.

The formula for velocity is:

Velocity = (# Using Samples * 100) / (# Using Samples + # Delay Samples)
Could you please let me know what is the meaning of "samples" in the above citation.

Thanks & Regards,

vasanthz · Posted: Wed Jul 31, 2013 2:59 pm

I started reading a material and the very first sentence is "Performance Management on an MVS system is a difficult and demanding task"
I was like, "ya very encouraging.. Thanks Mr.Author" :-)

Robert Sample · Posted: Wed Jul 31, 2013 4:39 pm

If you don't have the Redbook System Programmer's Guide to: Workload Management (SG24-6472), you need it. Page 14 of the Redbook includes

vasanthz · Posted: Wed Jul 31, 2013 4:53 pm

Hello Robert,
Thank you for your input.

I can now understand WLM does sampling, but still unable to understand the formula,
Velocity = (# Using Samples * 100) / (# Using Samples + # Delay Samples)

I would get the WLM guide and have a start from there.

started with "RMF performance management guide" and to understand it, I had to go to "RMF user guide" and now to understand RMF, I have to go to "WLM user guide"

Regards,

Robert Sample · Posted: Wed Jul 31, 2013 5:08 pm

Job X is running in an address space. Out of the first 100 times WLM samples while job X is executing, job X is using CPU on 43 of those and waiting (for I/O, for example) for 57 of them. The velocity will be 43. Velocity is basically a measure of how active the job / service class / TSO user / ... is within the system.

Velocity is used in setting goals for WLM -- service class PRDBATMD for example may have a velocity goal of 20. This means all jobs executing in that service class (whether 1 or 40) will have WLM attempting to give 20% of the processing time to them.

A couiple of notes to keep in mind while looking at WLM:
1. WLM is goals; PI (Performance Index) measures how well WLM is doing at achieving those goals
2. WLM performance is really only effective when CPU is 100%; if the CPU utilization is 80% then the address spaces on the system are not using all the CPU anyway.

vasanthz · Posted: Wed Jul 31, 2013 6:33 pm

Robert,
Thanks a lot for the layman example, Im able to understand it.

gratefully,

vasanthz · Posted: Thu Aug 08, 2013 4:15 pm

Hello,
I read two separate sentences, which I am trying to co-relate and understand

"Beginning in z/OS V1R3 systems will IPL and run in WLM goal mode only"

"In a system that is not set up in Goal mode, the system is controlled by performance groups and domains, as specified in IEAICS and IEAIPS members."

Does this mean that the IEAICS and IEAIPS method of managing workload are outdated techniques and no longer followed?

Is WLM goal mode(or some competitive software) omnipresent in all zOS 1.3 and later till 1.13 for managing workloads?

Thanks & Regards,

Robert Sample · Posted: Thu Aug 08, 2013 5:53 pm

There may still be a few sites running very old releases of z/OS or OS/390, in which case they may not be using WLM goal mode. However, since z/OS 1.3 the use of WLM has not been optional -- the old method of setting up performance groups and domains via IEAICSxx AND IEAIPSxx won't work with WLM goal mode.

WLM is present in all releases of z/OS as well as some releases of OS/390 (IIRC), so it has been around a while. As far as I know, there are no competitor products (although I've never looked) especially since WLM is available for all z/OS systems anyway. And setting up goals under WLM with the ISPF panels is much easier than translating performance metrics into IEAICSxx AND IEAIPSxx parameters.

vasanthz · Posted: Thu Aug 08, 2013 6:20 pm

Thank you Robert. Makes sense