[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20230215182547.GA17571@blackbody.suse.cz>
Date: Wed, 15 Feb 2023 19:25:48 +0100
From: Michal Koutný <mkoutny@...e.com>
To: Kairui Song <ryncsn@...il.com>
Cc: Johannes Weiner <hannes@...xchg.org>,
Suren Baghdasaryan <surenb@...gle.com>,
Chengming Zhou <zhouchengming@...edance.com>,
Tejun Heo <tj@...nel.org>, Ingo Molnar <mingo@...hat.com>,
Peter Zijlstra <peterz@...radead.org>, cgroups@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/2] sched/psi: iterate through cgroups directly
On Thu, Feb 16, 2023 at 01:49:22AM +0800, Kairui Song <ryncsn@...il.com> wrote:
> With rootcg:
> 55718.9 op/s (unpatched) compared to 55862.2 (patched)
> With 5 levels:
> 49975.5 op/s (unpatched) compared to 50778.5 op/s (patched)
>
> Previous tests are a bit biased since I only run the test for 100 * 3
> times, or maybe it is sensitive to some random kernel structure
> changes.
>
> But I ignored one important thing in my previous test, that the
> performance actually drops heavily with deeper levers of cgroups:
> With 8 levels:
> 48902.4 op/s (unpatched) compared to 47476.6 op/s (patched)
> With 50 levels of cgroup:
> 25605.7375 op/s (unpatched) compared to 20417.275 op/s (patched)
IIUC, one could also interpret this as the parent caching within
psi_group is effective especially with deep hierarchies.
I'd say practical hierarchies are below 10 levels deep. But yeah, the
averaged results aren't so impressive.
Thanks for sharing your insights,
Michal
Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)
Powered by blists - more mailing lists