lkml.org 
[lkml]   [2022]   [Jun]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4] sched/fair: Introduce SIS_UTIL to search idle CPU based on sum of util_avg
On Fri, Jun 24, 2022 at 09:29:58AM +0200, Peter Zijlstra wrote:
> On Wed, Jun 22, 2022 at 12:06:55PM +0530, K Prateek Nayak wrote:
> > Hello Chenyu,
> >
> > I'm sorry for the delay. The testing took a while but below are
> > the results from testing on our system.
> >
> > tl;dr
> >
> > o We ran all the tests with with SIS_PROP disabled.
> > o tbench reaches close to saturation early with 256 clients.
> > o schbench shows improvements for low worker counts.
> > o All other benchmark results seem comparable to tip.
> > We don't see any serious regressions with v4.
> >
> > > @@ -61,6 +61,7 @@ SCHED_FEAT(TTWU_QUEUE, true)
> > > * When doing wakeups, attempt to limit superfluous scans of the LLC domain.
> > > */
> > > SCHED_FEAT(SIS_PROP, true)
> >
> > SIS_PROP was disabled in our testing as follows:
> >
> > --
> > -SCHED_FEAT(SIS_PROP, true)
> > +SCHED_FEAT(SIS_PROP, false)
>
> So how about I make this change.
>
> > With v4 on the current tip, I don't see any need for
> > a special case for systems with smaller LLCs with
> > SIS_PROP disabled and SIS_UITL enable. Even SIS Efficiency
> > seems to be better with SIS_UTIL for hackbench.
> >
> > Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
>
> And apply this thing, lets see how it fares..

OK, thanks, Peter.


Best,
Chenyu

\
 
 \ /
  Last update: 2022-06-24 14:20    [W:0.053 / U:2.748 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site