From: Patrick Bellasi <patrick.bellasi@arm.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Vincent Guittot <vincent.guittot@linaro.org>,
Ingo Molnar <mingo@kernel.org>,
linux-kernel <linux-kernel@vger.kernel.org>,
"Rafael J. Wysocki" <rjw@rjwysocki.net>,
Juri Lelli <juri.lelli@redhat.com>,
Dietmar Eggemann <dietmar.eggemann@arm.com>,
Morten Rasmussen <Morten.Rasmussen@arm.com>,
viresh kumar <viresh.kumar@linaro.org>,
Valentin Schneider <valentin.schneider@arm.com>,
Quentin Perret <quentin.perret@arm.com>
Subject: Re: [PATCH v5 00/10] track CPU utilization
Date: Tue, 5 Jun 2018 16:38:26 +0100 [thread overview]
Message-ID: <20180605153826.GE32302@e110439-lin> (raw)
In-Reply-To: <20180605141809.GV12180@hirez.programming.kicks-ass.net>
On 05-Jun 16:18, Peter Zijlstra wrote:
> On Mon, Jun 04, 2018 at 08:08:58PM +0200, Vincent Guittot wrote:
> > On 4 June 2018 at 18:50, Peter Zijlstra <peterz@infradead.org> wrote:
>
> > > So this patch-set tracks the !cfs occupation using the same function,
> > > which is all good. But what, if instead of using that to compensate the
> > > OPP selection, we employ that to renormalize the util signal?
> > >
> > > If we normalize util against the dynamic (rt_avg affected) cpu_capacity,
> > > then I think your initial problem goes away. Because while the RT task
> > > will push the util to .5, it will at the same time push the CPU capacity
> > > to .5, and renormalized that gives 1.
And would not that mean also that a 50% task co-scheduled with the
same 50% RT task, will be reported as a 100% util_avg task?
> > >
> > > NOTE: the renorm would then become something like:
> > > scale_cpu = arch_scale_cpu_capacity() / rt_frac();
>
> Should probably be:
>
> scale_cpu = atch_scale_cpu_capacity() / (1 - rt_frac())
>
> > >
> > >
> > > On IRC I mentioned stopping the CFS clock when preempted, and while that
> > > would result in fixed numbers, Vincent was right in pointing out the
> > > numbers will be difficult to interpret, since the meaning will be purely
> > > CPU local and I'm not sure you can actually fix it again with
> > > normalization.
> > >
> > > Imagine, running a .3 RT task, that would push the (always running) CFS
> > > down to .7, but because we discard all !cfs time, it actually has 1. If
> > > we try and normalize that we'll end up with ~1.43, which is of course
> > > completely broken.
> > >
> > >
> > > _However_, all that happens for util, also happens for load. So the above
> > > scenario will also make the CPU appear less loaded than it actually is.
> >
> > The load will continue to increase because we track runnable state and
> > not running for the load
>
> Duh yes. So renormalizing it once, like proposed for util would actually
> do the right thing there too. Would not that allow us to get rid of
> much of the capacity magic in the load balance code?
>
> /me thinks more..
>
> Bah, no.. because you don't want this dynamic renormalization part of
> the sums. So you want to keep it after the fact. :/
>
> > As you mentioned, scale_rt_capacity give the remaining capacity for
> > cfs and it will behave like cfs util_avg now that it uses PELT. So as
> > long as cfs util_avg < scale_rt_capacity(we probably need a margin)
> > we keep using dl bandwidth + cfs util_avg + rt util_avg for selecting
> > OPP because we have remaining spare capacity but if cfs util_avg ==
> > scale_rt_capacity, we make sure to use max OPP.
What will happen for the 50% task of the example above?
> Good point, when cfs-util < cfs-cap then there is idle time and the util
> number is 'right', when cfs-util == cfs-cap we're overcommitted and
> should go max.
Again I cannot easily read the example above...
Would that mean that a 50% CFS task, preempted by a 50% RT task (which
already set OPP to max while RUNNABLE) will end up running at the max
OPP too?
> Since the util and cap values are aligned that should track nicely.
True... the only potential issue I see is that we are steering PELT
behaviors towards better driving schedutil to run high-demand
workloads while _maybe_ affecting quite sensibly the capacity of PELT
to describe how much CPU a task uses.
Ultimately, utilization has always been a metric on "how much you
use"... while here it seems to me we are bending it to be something to
define "how fast you have to run".
--
#include <best/regards.h>
Patrick Bellasi
next prev parent reply other threads:[~2018-06-05 15:38 UTC|newest]
Thread overview: 99+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-05-25 13:12 [PATCH v5 00/10] track CPU utilization Vincent Guittot
2018-05-25 13:12 ` [PATCH v5 01/10] sched/pelt: Move pelt related code in a dedicated file Vincent Guittot
2018-05-25 14:26 ` Quentin Perret
2018-05-25 16:14 ` Peter Zijlstra
2018-05-29 8:21 ` Quentin Perret
2018-05-25 18:04 ` Patrick Bellasi
2018-05-29 14:55 ` Quentin Perret
2018-05-29 15:02 ` Vincent Guittot
2018-05-29 15:04 ` Quentin Perret
2018-05-25 13:12 ` [PATCH v5 02/10] sched/rt: add rt_rq utilization tracking Vincent Guittot
2018-05-25 15:54 ` Patrick Bellasi
2018-05-29 13:29 ` Vincent Guittot
2018-05-30 9:32 ` Patrick Bellasi
2018-05-30 10:06 ` Vincent Guittot
2018-05-30 11:01 ` Patrick Bellasi
2018-05-30 14:39 ` Vincent Guittot
2018-05-25 13:12 ` [PATCH v5 03/10] cpufreq/schedutil: add rt " Vincent Guittot
2018-05-30 7:03 ` Viresh Kumar
2018-05-30 8:23 ` Vincent Guittot
2018-05-30 9:40 ` Patrick Bellasi
2018-05-30 9:53 ` Vincent Guittot
2018-05-30 16:46 ` Quentin Perret
2018-05-31 8:46 ` Juri Lelli
2018-06-01 16:23 ` Peter Zijlstra
2018-06-01 17:23 ` Patrick Bellasi
2018-06-04 10:17 ` Quentin Perret
2018-06-04 15:16 ` Patrick Bellasi
2018-05-25 13:12 ` [PATCH v5 04/10] sched/dl: add dl_rq " Vincent Guittot
2018-05-30 10:50 ` Patrick Bellasi
2018-05-30 11:51 ` Vincent Guittot
2018-05-25 13:12 ` [PATCH v5 05/10] cpufreq/schedutil: get max utilization Vincent Guittot
2018-05-28 10:12 ` Juri Lelli
2018-05-28 14:57 ` Vincent Guittot
2018-05-28 15:22 ` Juri Lelli
2018-05-28 16:34 ` Vincent Guittot
2018-05-31 10:27 ` Patrick Bellasi
2018-05-31 13:02 ` Vincent Guittot
2018-06-01 13:53 ` Vincent Guittot
2018-06-01 17:45 ` Joel Fernandes
2018-06-04 6:41 ` Vincent Guittot
2018-06-04 7:04 ` Juri Lelli
2018-06-04 7:14 ` Vincent Guittot
2018-06-04 10:12 ` Juri Lelli
2018-06-04 12:35 ` Vincent Guittot
2018-05-29 5:08 ` Joel Fernandes
2018-05-29 6:31 ` Juri Lelli
2018-05-29 6:48 ` Vincent Guittot
2018-05-29 9:47 ` Juri Lelli
2018-05-29 8:40 ` Quentin Perret
2018-05-29 9:52 ` Juri Lelli
2018-05-30 8:37 ` Quentin Perret
2018-05-30 8:51 ` Juri Lelli
2018-05-25 13:12 ` [PATCH v5 06/10] sched: remove rt and dl from sched_avg Vincent Guittot
2018-05-25 13:12 ` [PATCH v5 07/10] sched/irq: add irq utilization tracking Vincent Guittot
2018-05-30 15:55 ` Dietmar Eggemann
2018-05-30 18:45 ` Vincent Guittot
2018-05-31 16:54 ` Dietmar Eggemann
2018-06-06 16:06 ` Vincent Guittot
2018-06-07 8:29 ` Dietmar Eggemann
2018-06-07 8:44 ` Vincent Guittot
2018-06-07 9:06 ` Dietmar Eggemann
2018-05-25 13:12 ` [PATCH v5 08/10] cpufreq/schedutil: take into account interrupt Vincent Guittot
2018-05-28 10:41 ` Juri Lelli
2018-05-28 12:06 ` Vincent Guittot
2018-05-28 12:37 ` Juri Lelli
2018-05-25 13:12 ` [PATCH v5 09/10] sched: remove rt_avg code Vincent Guittot
2018-05-25 13:12 ` [PATCH v5 10/10] proc/sched: remove unused sched_time_avg_ms Vincent Guittot
2018-06-04 16:50 ` [PATCH v5 00/10] track CPU utilization Peter Zijlstra
2018-06-04 17:13 ` Quentin Perret
2018-06-04 18:08 ` Vincent Guittot
2018-06-05 14:18 ` Peter Zijlstra
2018-06-05 15:03 ` Juri Lelli
2018-06-05 15:38 ` Patrick Bellasi [this message]
2018-06-05 22:27 ` Peter Zijlstra
2018-06-06 9:44 ` Quentin Perret
2018-06-06 9:59 ` Vincent Guittot
2018-06-06 10:02 ` Vincent Guittot
2018-06-06 10:12 ` Quentin Perret
2018-06-05 8:36 ` Vincent Guittot
2018-06-05 10:57 ` Quentin Perret
2018-06-05 11:59 ` Vincent Guittot
2018-06-05 13:12 ` Quentin Perret
2018-06-05 13:18 ` Vincent Guittot
2018-06-05 13:52 ` Quentin Perret
2018-06-05 13:55 ` Vincent Guittot
2018-06-05 14:09 ` Quentin Perret
2018-06-05 14:21 ` Quentin Perret
2018-06-05 12:11 ` Juri Lelli
2018-06-05 13:05 ` Quentin Perret
2018-06-05 13:15 ` Juri Lelli
2018-06-05 14:01 ` Quentin Perret
2018-06-05 14:13 ` Juri Lelli
2018-06-06 13:05 ` Claudio Scordino
2018-06-06 13:20 ` Quentin Perret
2018-06-06 13:53 ` Claudio Scordino
2018-06-06 14:10 ` Quentin Perret
2018-06-06 21:05 ` luca abeni
2018-06-07 8:25 ` Quentin Perret
2018-06-06 20:53 ` luca abeni
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20180605153826.GE32302@e110439-lin \
--to=patrick.bellasi@arm.com \
--cc=Morten.Rasmussen@arm.com \
--cc=dietmar.eggemann@arm.com \
--cc=juri.lelli@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@kernel.org \
--cc=peterz@infradead.org \
--cc=quentin.perret@arm.com \
--cc=rjw@rjwysocki.net \
--cc=valentin.schneider@arm.com \
--cc=vincent.guittot@linaro.org \
--cc=viresh.kumar@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).