linux-kernel - Re: kernel/scheduler: The Linux scheduler doesn't scale to more than 8 cores?

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 2 Nov 2021 17:04:02 +0100
From:   Peter Zijlstra <peterz@...radead.org>
To:     Vincent Guittot <vincent.guittot@...aro.org>
Cc:     "Morotti, Romain (London)" <romain.morotti@....com>,
        Ingo Molnar <mingo@...hat.com>,
        Juri Lelli <juri.lelli@...hat.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: kernel/scheduler: The Linux scheduler doesn't scale to more than
 8 cores?

On Tue, Nov 02, 2021 at 04:25:05PM +0100, Vincent Guittot wrote:
> > From the first line, the CPU count is capped to 8.
> >
> > Thus the scheduler scaling won't be more than a factor of 4, no matter how many CPUs there are. (1+log2(8))
> >
> > There's also a linear scaling option, that is similarly capped to 8 times, because 8 CPUs.
> 
> That's a good point and I never realized this limitation in the scaling factor.
> 
> Peter mentioned that there were interactivity problems with large
> scale factors at the time it was added. But the scheduler has changed
> since and it would be interesting to run benchmarks on more recent
> platforms with larger factor

The important note here is that this limit is about timescales, not
anything else. Previously a 64 cpu system would blow up the latency to
6ms * (1 + ln)2(64)) = 42ms, which is atrocious.