A few years back, I did a deep dive into tuning CPU usage across a multitude of very small z/OS guests under z/VM, and WLM was certainly a big hitter for many of them, but as there were so many instances, I was able to see notable differences in WLM use between "LPARs", which was obviously "of interest". The upshot seemed to be that WLM costs had a fairly firm relationship with the number of active address spaces on the "LPAR", presumably down to the amount of sampling that WLM has to do against each address space every 250ms (I think). I did enquire of IBM as to whether the sampling rate could be "adjusted", and that came back with a negative response (not really a surprise). So the obvious answer may be to only have address spaces started, when they are only really needed to be there. Although you may need to assess the cost of stopping/starting those address spaces, versus the background WLM cost.
On Mon, 4 Mar 2024 at 23:08, Wendell Lovewell < 000001e9c0ee0673-dmarc-requ...@listserv.ua.edu> wrote: > This is probably a strange question, but is there a way to reduce WLM cpu > usage? Here's the situation: > > - The system is a lightly used development system. Unless something is in > a loop (very rare), CPU % probably is usually less than 10%. And except > for system regions & CICS, it's rare to have multiple jobs running > concurrently. > - Only one processor defined to the VM. No ZIIP either. > - We are charged for CPU cycles. > - WLM is the highest consumer of CPU. JES2, TCPIP, ZFS and SDSFAUX round > out the top 5 consumers. > > There is a lot of information about WLM tuning, but as best I can tell > almost none of it relates to reducing WLM usage. > > From reading the Init & Tuning manual, I'm trying these settings: > AIMANAGEMENT=NO > HIPERDISPATCH=NO > CCCAWMT=450000 > RMPTTOM=15000 > > I was thinking that perhaps reducing whatever processing intervals I could > might help. But I can't tell these changes made a difference. (I don't > have a tool to measure WLM usage.) > > Any suggestions would be appreciated. > > TIA, > > Wendell > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
