Thanks Alfred, Here are the usual benchmarks. I've also experimented with turbostat. It's seems the frequency scaling with schedutil governor is working correctly. The energy spent while the load is low (watching video) is a little higher than with CFS though.
Sorry, for having been so impatient... I have the "v5.13-prjc-r1" running for 17h now. Like earlier, I use it with PDS and default "sched_timeslice" and it works great, meaning, that some CPU overhead is reduced indeed (subjective experience and gkrellm CPU observation in my usual use cases).
Thanks Alfred,
ReplyDeleteHere are the usual benchmarks.
I've also experimented with turbostat. It's seems the frequency scaling with schedutil governor is working correctly.
The energy spent while the load is low (watching video) is a little higher than with CFS though.
https://docs.google.com/spreadsheets/d/163U3H-gnVeGopMrHiJLeEY1b7XlvND2yoceKbOvQRm4/edit#gid=1819453664
Pedro
Thanks for your tesing, Pddro.
DeleteSo far, prjc(BMQ) scheduler looks very balance, IMO. :)
Compiled and ran the kernel in i7@work and Ryzen@home, so far so good.
ReplyDeleteThanks Alfred!
BR,
Eduardo
I was wondering if Project C is going to need some rework to support Alder Lake's big.LITTLE architecture?
ReplyDelete@Alfred:
ReplyDeleteSomehow I'm missing an announcement about the v5.13 releases here on the blog.
Is the code safe to use or is there any reason for it?
Anyway, thank you very much for your work!
BR,
Manuel
Sorry, for having been so impatient...
DeleteI have the "v5.13-prjc-r1" running for 17h now. Like earlier, I use it with PDS and default "sched_timeslice" and it works great, meaning, that some CPU overhead is reduced indeed (subjective experience and gkrellm CPU observation in my usual use cases).
Thank you very much,
Manuel