[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Z9kBzeGAnxtQIvi6@gpd3>
Date: Tue, 18 Mar 2025 06:17:01 +0100
From: Andrea Righi <arighi@...dia.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Tejun Heo <tj@...nel.org>, Joel Fernandes <joelagnelf@...dia.com>,
linux-kernel@...r.kernel.org, David Vernet <void@...ifault.com>,
Changwoo Min <changwoo@...lia.com>, Ingo Molnar <mingo@...hat.com>,
Juri Lelli <juri.lelli@...hat.com>,
Vincent Guittot <vincent.guittot@...aro.org>,
Dietmar Eggemann <dietmar.eggemann@....com>,
Steven Rostedt <rostedt@...dmis.org>,
Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>,
Valentin Schneider <vschneid@...hat.com>
Subject: Re: [PATCH RFC] sched_ext: Choose prev_cpu if idle and cache affine
without WF_SYNC
On Mon, Mar 17, 2025 at 06:44:31PM +0100, Peter Zijlstra wrote:
> On Mon, Mar 17, 2025 at 06:30:57PM +0100, Andrea Righi wrote:
>
> > I guess the question is: what is more expensive in general on task wakeup?
> > 1) a cross-node migration or 2) running on a partially busy SMT core?
>
> That totally depends on both the workload and the actual machine :/
>
> If you have 'fast' numa and not a very big footprint, the numa
> migrations aren't too bad. OTOH if you have sucky numa or your memory
> footprint is significant, then running on the wrong node is super
> painful.
Right, there isn't a single "best solution" in general, I guess we just
need to pick what we think it works best in most cases, set that as
default, and then leave it to the BPF schedulers to adjust the behavior as
needed (at the end this is just a default policy).
Thanks,
-Andrea
Powered by blists - more mailing lists