[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 02 Feb 2018 18:18:45 +0100
From: Mike Galbraith <efault@....de>
To: Sergio Lopez <slp@...rega.org>, linux-kernel@...r.kernel.org
Cc: mingo@...hat.com, peterz@...radead.org
Subject: Re: [RFC PATCH] sched/fair: Optionally restrict SMT scheduling
("restrict_smt")
On Fri, 2018-02-02 at 16:14 +0100, Sergio Lopez wrote:
> On Thu, 2018-02-01 at 15:37 +0100, Mike Galbraith wrote:
> > On Thu, 2018-02-01 at 13:37 +0100, Sergio Lopez wrote:
> > > Compared against "isolcpus", "restrict_smt" presents the following
> > > advantages:
> > > - Doesn't require prior knowledge of the CPU topology, making it
> > > more friendly to both end users and automation tools.
> > > - Load balancing is still active for tasks pinned to multiple
> > > secondary SMT threads.
> >
> > This says that everybody and everything in the box is way smarter
> > than the admin.
>
> If that's an argument in favor of manual pinning and isolation, I have
> to disagree with you. That's an unnecessary burden put on the shoulders
> of the admin.
Give the admin job to one of those users you mentioned who knows which
threads are "secondary" threads, he/she won't be burdened at all.
-Mike
Powered by blists - more mailing lists