From: Valentin Schneider <valentin.schneider@arm.com>
To: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Ingo Molnar <mingo@redhat.com>,
Peter Zijlstra <peterz@infradead.org>,
Juri Lelli <juri.lelli@redhat.com>,
Dietmar Eggemann <dietmar.eggemann@arm.com>,
Steven Rostedt <rostedt@goodmis.org>,
Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
linux-kernel <linux-kernel@vger.kernel.org>,
Phil Auld <pauld@redhat.com>, Parth Shah <parth@linux.ibm.com>,
Hillf Danton <hdanton@sina.com>
Subject: Re: [PATCH v3 4/5] sched/pelt: Add a new runnable average signal
Date: Mon, 24 Feb 2020 15:57:36 +0000 [thread overview]
Message-ID: <e4def999-b784-6b26-3748-a2ad57f79c6d@arm.com> (raw)
In-Reply-To: <CAKfTPtAQ_09wVM7zjrHDB+gJXpb5OH6CBvfKC7OB_Bo0Hd41vA@mail.gmail.com>
I somehow lost track of that email, sorry for the delayed response.
On 2/21/20 8:56 AM, Vincent Guittot wrote:
> On Thu, 20 Feb 2020 at 17:11, Valentin Schneider
> <valentin.schneider@arm.com> wrote:
>>
>> On 20/02/2020 14:36, Vincent Guittot wrote:
>>> I agree that setting by default to SCHED_CAPACITY_SCALE is too much
>>> for little core.
>>> The problem for little core can be fixed by using the cpu capacity instead
>>>
>>
>> So that's indeed better for big.LITTLE & co. Any reason however for not
>> aligning with the initialization of util_avg ?
>
> The runnable_avg is the unweighted version of the load_avg so they
> should both be sync at init and SCHED_CAPACITY_SCALE is in fact the
> right value. Using cpu_scale is the same for smp and big core so we
> can use it instead.
>
> Then, the initial value of util_avg has never reflected some kind of
> realistic value for the utilization of a new task, especially if those
> tasks will become big ones. Runnable_avg now balances this effect to
> say that we don't know what will be the behavior of the new task,
> which might end up using all spare capacity although current
> utilization is low and CPU is not "fully used".
I'd argue that the init values we pick for either runnable_avg or util_avg
are both equally bogus.
> In fact, this is
> exactly the purpose of runnable: highlight that there is maybe no
> spare capacity even if CPU's utilization is low because of external
> event like task migration or having new tasks with most probably wrong
> utilization.
>
> That being said, there is a bigger problem with the current version of
> this patch, which is that I forgot to use runnable in
> update_sg_wakeup_stats(). I have a patch that fixes this problem.
>
> Also, I have tested both proposals with hackbench on my octo cores and
> using cpu_scale gives slightly better results than util_avg, which
> probably reflects the case I mentioned above.
>
> grp cpu_scale util_avg improvement
> 1 1,191(+/-0.77 %) 1,204(+/-1.16 %) -1.07 %
> 4 1,147(+/-1.14 %) 1,195(+/-0.52 %) -4.21 %
> 8 1,112(+/-1,52 %) 1,124(+/-1,45 %) -1.12 %
> 16 1,163(+/-1.72 %) 1,169(+/-1.58 %) -0,45 %
>
Interesting, thanks for providing the numbers. I'd be curious to figure out
where the difference really stems from, but in the meantime consider me
convinced ;)
next prev parent reply other threads:[~2020-02-24 15:57 UTC|newest]
Thread overview: 49+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-02-14 15:27 [PATCH v2 0/5] remove runnable_load_avg and improve group_classify Vincent Guittot
2020-02-14 15:27 ` [PATCH v2 1/5] sched/fair: Reorder enqueue/dequeue_task_fair path Vincent Guittot
2020-02-18 12:37 ` Dietmar Eggemann
2020-02-18 13:22 ` Peter Zijlstra
2020-02-18 14:15 ` Vincent Guittot
2020-02-19 11:07 ` Dietmar Eggemann
2020-02-19 16:26 ` Vincent Guittot
2020-02-20 13:38 ` Dietmar Eggemann
[not found] ` <20200222152541.GA11669@geo.homenetwork>
2020-02-26 16:30 ` Vincent Guittot
2020-02-14 15:27 ` [PATCH v2 2/5] sched/numa: Replace runnable_load_avg by load_avg Vincent Guittot
2020-02-18 12:37 ` Dietmar Eggemann
2020-02-18 13:50 ` Mel Gorman
2020-02-18 14:17 ` Vincent Guittot
2020-02-18 14:42 ` Dietmar Eggemann
2020-02-18 14:54 ` Valentin Schneider
2020-02-18 15:33 ` Vincent Guittot
2020-02-18 15:38 ` Mel Gorman
2020-02-18 16:50 ` Valentin Schneider
2020-02-18 17:41 ` Mel Gorman
2020-02-18 17:54 ` Valentin Schneider
2020-02-18 16:51 ` Vincent Guittot
2020-02-14 15:27 ` [PATCH v2 3/5] sched/pelt: Remove unused runnable load average Vincent Guittot
2020-02-21 9:57 ` Dietmar Eggemann
2020-02-21 11:56 ` Vincent Guittot
2020-02-14 15:27 ` [PATCH v2 4/5] sched/pelt: Add a new runnable average signal Vincent Guittot
2020-02-18 14:54 ` Valentin Schneider
2020-02-18 15:12 ` Peter Zijlstra
2020-02-18 15:28 ` Vincent Guittot
2020-02-18 16:30 ` Valentin Schneider
2020-02-18 21:19 ` Valentin Schneider
2020-02-19 9:02 ` Vincent Guittot
2020-02-19 9:08 ` Mel Gorman
2020-02-19 12:55 ` [PATCH v3 " Vincent Guittot
2020-02-19 14:02 ` Mel Gorman
2020-02-19 20:10 ` Valentin Schneider
2020-02-20 14:36 ` Vincent Guittot
2020-02-20 16:11 ` Valentin Schneider
2020-02-21 8:56 ` Vincent Guittot
2020-02-24 15:57 ` Valentin Schneider [this message]
2020-02-21 9:04 ` Mel Gorman
2020-02-21 9:25 ` Vincent Guittot
2020-02-21 10:40 ` Mel Gorman
2020-02-21 13:28 ` Vincent Guittot
2020-02-20 15:04 ` Dietmar Eggemann
2020-02-21 9:44 ` Dietmar Eggemann
2020-02-21 11:47 ` Vincent Guittot
2020-02-14 15:27 ` [PATCH v2 5/5] sched/fair: Take into account runnable_avg to classify group Vincent Guittot
2020-02-15 21:58 ` [PATCH v2 0/5] remove runnable_load_avg and improve group_classify Mel Gorman
2020-02-21 9:58 ` Dietmar Eggemann
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=e4def999-b784-6b26-3748-a2ad57f79c6d@arm.com \
--to=valentin.schneider@arm.com \
--cc=bsegall@google.com \
--cc=dietmar.eggemann@arm.com \
--cc=hdanton@sina.com \
--cc=juri.lelli@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mgorman@suse.de \
--cc=mingo@redhat.com \
--cc=parth@linux.ibm.com \
--cc=pauld@redhat.com \
--cc=peterz@infradead.org \
--cc=rostedt@goodmis.org \
--cc=vincent.guittot@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).