Re: [PATCH 10/10] sched/fair: Implement an EEVDF like policy

From: Mike Galbraith
Date: Sat Mar 11 2023 - 00:55:16 EST


On Fri, 2023-03-10 at 21:38 +0100, Peter Zijlstra wrote:
> On Thu, Mar 09, 2023 at 05:23:33AM +0100, Mike Galbraith wrote:
> > Ok, seems there must be a math booboo lurking.
>
> Yep that.. please try the version I pushed out here:
>
>   https://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git/log/?h=sched/eevdf
>
> (It's based on 6.3-rc1, but it should be trivial to rebase onto v6.2 if
> you so wish.)

I stuffed it into .today. Yup, extreme sleeper stinginess is gone.

tbench/netperf numbers look roughly as previously. massive_intr vs
desktop CPU distribution improved as expected, but (to me) oddly, the
amount of desktop beating up itself did not.

These are all .today

perf.data.eevdf
Task | Runtime ms | Switches | Avg delay ms | Max delay ms | Sum delay ms | Max delay start | Max delay end |
massive_intr:(9) |1499451.694 ms | 647385 | avg: 0.940 ms | max: 27.969 ms | sum:608348.424 ms | max start: 467.219496 s | max end: 467.247465 s
TOTAL: |2369880.457 ms | 2797272 | | 1243484.857 ms |
perf.data.eevdf_FAIR_SLEEPERS
Task | Runtime ms | Switches | Avg delay ms | Max delay ms | Sum delay ms | Max delay start | Max delay end |
massive_intr:(9) |1415743.272 ms | 714999 | avg: 1.013 ms | max: 73.722 ms | sum:724624.600 ms | max start: 2678.846216 s | max end: 2678.919938 s
TOTAL: |2362333.392 ms | 2871586 | | 1238748.802 ms |
perf.data.master
Task | Runtime ms | Switches | Avg delay ms | Max delay ms | Sum delay ms | Max delay start | Max delay end |
massive_intr:(9) |1463798.673 ms | 598383 | avg: 0.730 ms | max: 40.029 ms | sum:437069.674 ms | max start: 945.022965 s | max end: 945.062994 s
TOTAL: |2370011.937 ms | 1109828 | | 616408.789 ms |