[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <15851309-1c3b-4506-b094-fd445941f142@linux.ibm.com>
Date: Tue, 14 Jan 2025 19:13:51 +0100
From: Tobias Huschle <huschle@...ux.ibm.com>
To: Shrikanth Hegde <sshegde@...ux.ibm.com>
Cc: mingo@...hat.com, peterz@...radead.org, juri.lelli@...hat.com,
vincent.guittot@...aro.org, dietmar.eggemann@....com,
rostedt@...dmis.org, bsegall@...gle.com, mgorman@...e.de,
vschneid@...hat.com, linux-s390@...r.kernel.org,
linuxppc-dev@...ts.ozlabs.org, linux-kernel@...r.kernel.org
Subject: Re: [RFC PATCH 0/2] sched/fair: introduce new scheduler group type
group_parked
On 10/12/2024 21:24, Shrikanth Hegde wrote:
> On 12/9/24 13:35, Tobias Huschle wrote:
[...]
>
> It was happening with 100% stress-ng case. I was wondering since i dont
> have no-hz full enabled.
> I found out the reason why and one way to do is to trigger active load
> balance if there are any parked cpus
> in the group. That probably needs a IS_ENABLED check not to hurt the
> regular case.
>
> Also, I gave a try to include arch_cpu_parked in idle_cpu and friends.
> It seems to working for me.
> I will attach the code below. It simplifies code quite a bit.
>
> Also, I am thinking to rely on active balance codepath more than the
> regular pull model.
> so this would be akin to asym packing codepaths. The below code does
> that too.
>
> Feel free to take the bits as necessary if it works.
>
Thanks a lot for your your comments and proposals. I was working through
them and have a v2 almost ready. I'll be offline for the next 4 weeks
though and will post my v2 once I'm back.
[...]
Powered by blists - more mailing lists