Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1620496 > unrolled thread

Re: [PATCH v2] sched/fair: update scale invariance of PELT

Started byPeter Zijlstra <peterz@infradead.org>
First post2017-04-10 19:40 +0200
Last post2017-04-19 18:40 +0200
Articles 3 on this page of 23 — 2 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-10 19:40 +0200
    Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-11 10:00 +0200
      Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-11 11:00 +0200
        Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-11 11:50 +0200
          Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-11 12:50 +0200
            Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-11 13:00 +0200
              Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-11 15:10 +0200
                Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-12 13:30 +0200
                  Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-12 17:00 +0200
                    Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-12 17:50 +0200
                      Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-13 11:50 +0200
                  Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-13 15:40 +0200
                    Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-13 17:00 +0200
                      Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-13 20:10 +0200
                        Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-14 10:50 +0200
            Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-11 14:10 +0200
      Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-11 11:20 +0200
        Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-11 11:50 +0200
      Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-13 15:40 +0200
        Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-13 17:20 +0200
          Re: [PATCH v2] sched/fair: update scale invariance of PELT Peter Zijlstra <peterz@infradead.org> - 2017-04-13 18:20 +0200
            Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-14 11:00 +0200
              Re: [PATCH v2] sched/fair: update scale invariance of PELT Vincent Guittot <vincent.guittot@linaro.org> - 2017-04-19 18:40 +0200

Page 2 of 2 — ← Prev page 1 [2]

#1623160

FromPeter Zijlstra <peterz@infradead.org>
Date2017-04-13 18:20 +0200
Message-ID<tvPDs-5oA-9@gated-at.bofh.it>
In reply to#1623114
On Thu, Apr 13, 2017 at 05:16:20PM +0200, Vincent Guittot wrote:
> On 13 April 2017 at 15:39, Peter Zijlstra <peterz@infradead.org> wrote:

> > OK, so the reason util_avg varies is because we compute it wrong. And I
> > think we can easily fix that once we pull out all the factors (which
> > would mean your patch and the pulling out of weight patch which still
> > needs to be finished).
> 
> That would be great to remove this unwanted variation.

So the problem with the _avg stuff is that we include the d3 segment,
that is the unfinished current window. Since we only re-compute the _avg
whenever we roll over, the intent already seems to be to only compute it
on completed windows.

But because 'complicated/expensive', its hard to not include d3 and thus
we get the wobble.

Once we compute pure running/runnable sums, without extra contrib
factors, we can simply subtract our d3 term from sum when doing the
division and change the divider to LOAD_AVG_MAX*y, getting the stable
_avg over all completed windows.

(you could do the same with factors, but then we get to do a bunch of
extra multiplications which aren't free).

> >
> > But you're comparing against util_sum here, that behaves slightly
> > different. I think you want 'util_sum >= 1024 * (LOAD_AVG_MAX - 1024)'
> > instead.
> 
> yes, the variation happens on the util_sum

Well, for util_sum its simple to ignore the current window, which is
what the suggested equation does (note that LOAD_AVG_MAX*y ==
LOAD_AVG_MAX-1024).

[toc] | [prev] | [next] | [standalone]

#1623577

FromVincent Guittot <vincent.guittot@linaro.org>
Date2017-04-14 11:00 +0200
Message-ID<tw5fb-7mS-15@gated-at.bofh.it>
In reply to#1623160
On 13 April 2017 at 18:13, Peter Zijlstra <peterz@infradead.org> wrote:
> On Thu, Apr 13, 2017 at 05:16:20PM +0200, Vincent Guittot wrote:
>> On 13 April 2017 at 15:39, Peter Zijlstra <peterz@infradead.org> wrote:
>
>> > OK, so the reason util_avg varies is because we compute it wrong. And I
>> > think we can easily fix that once we pull out all the factors (which
>> > would mean your patch and the pulling out of weight patch which still
>> > needs to be finished).
>>
>> That would be great to remove this unwanted variation.
>
> So the problem with the _avg stuff is that we include the d3 segment,
> that is the unfinished current window. Since we only re-compute the _avg
> whenever we roll over, the intent already seems to be to only compute it
> on completed windows.

yes make sense

>
> But because 'complicated/expensive', its hard to not include d3 and thus
> we get the wobble.
>
> Once we compute pure running/runnable sums, without extra contrib
> factors, we can simply subtract our d3 term from sum when doing the
> division and change the divider to LOAD_AVG_MAX*y, getting the stable
> _avg over all completed windows.

I'm going to make it a try to check that it removes the variation i'm seeing

>
> (you could do the same with factors, but then we get to do a bunch of
> extra multiplications which aren't free).
>
>> >
>> > But you're comparing against util_sum here, that behaves slightly
>> > different. I think you want 'util_sum >= 1024 * (LOAD_AVG_MAX - 1024)'
>> > instead.
>>
>> yes, the variation happens on the util_sum
>
> Well, for util_sum its simple to ignore the current window, which is
> what the suggested equation does (note that LOAD_AVG_MAX*y ==
> LOAD_AVG_MAX-1024).

[toc] | [prev] | [next] | [standalone]

#1626498

FromVincent Guittot <vincent.guittot@linaro.org>
Date2017-04-19 18:40 +0200
Message-ID<ty0O6-5Ir-25@gated-at.bofh.it>
In reply to#1623577
On 14 April 2017 at 10:49, Vincent Guittot <vincent.guittot@linaro.org> wrote:
> On 13 April 2017 at 18:13, Peter Zijlstra <peterz@infradead.org> wrote:
>> On Thu, Apr 13, 2017 at 05:16:20PM +0200, Vincent Guittot wrote:
>>> On 13 April 2017 at 15:39, Peter Zijlstra <peterz@infradead.org> wrote:
>>
>>> > OK, so the reason util_avg varies is because we compute it wrong. And I
>>> > think we can easily fix that once we pull out all the factors (which
>>> > would mean your patch and the pulling out of weight patch which still
>>> > needs to be finished).
>>>
>>> That would be great to remove this unwanted variation.
>>
>> So the problem with the _avg stuff is that we include the d3 segment,
>> that is the unfinished current window. Since we only re-compute the _avg
>> whenever we roll over, the intent already seems to be to only compute it
>> on completed windows.
>
> yes make sense
>
>>
>> But because 'complicated/expensive', its hard to not include d3 and thus
>> we get the wobble.
>>
>> Once we compute pure running/runnable sums, without extra contrib
>> factors, we can simply subtract our d3 term from sum when doing the
>> division and change the divider to LOAD_AVG_MAX*y, getting the stable
>> _avg over all completed windows.
>
> I'm going to make it a try to check that it removes the variation i'm seeing

I have sent a patchset based on your proposal that fix this variation issue

>
>>
>> (you could do the same with factors, but then we get to do a bunch of
>> extra multiplications which aren't free).
>>
>>> >
>>> > But you're comparing against util_sum here, that behaves slightly
>>> > different. I think you want 'util_sum >= 1024 * (LOAD_AVG_MAX - 1024)'
>>> > instead.
>>>
>>> yes, the variation happens on the util_sum
>>
>> Well, for util_sum its simple to ignore the current window, which is
>> what the suggested equation does (note that LOAD_AVG_MAX*y ==
>> LOAD_AVG_MAX-1024).

[toc] | [prev] | [standalone]

Page 2 of 2 — ← Prev page 1 [2]

Back to top | Article view | linux.kernel

csiph-web