linux-kernel - Re: [PATCH] mm: zswap: limit number of zpools based on CPU and RAM

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]

Message-ID: <CAKEwX=PBcxsCK+qRvsPLRnDobttzm75YuS6Js7Y4R1MKSf_RxQ@mail.gmail.com>
Date: Fri, 7 Jun 2024 10:26:02 +0100
From: Nhat Pham <nphamcs@...il.com>
To: Takero Funaki <flintglass@...il.com>
Cc: Johannes Weiner <hannes@...xchg.org>, Yosry Ahmed <yosryahmed@...gle.com>, 
	Chengming Zhou <chengming.zhou@...ux.dev>, Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org, 
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm: zswap: limit number of zpools based on CPU and RAM

On Thu, Jun 6, 2024 at 5:53 PM Takero Funaki <flintglass@...il.com> wrote:
>
> This patch limits the number of zpools used by zswap on smaller systems.
>
> Currently, zswap allocates 32 pools unconditionally. This was
> implemented to reduce contention on per-zpool locks. However, it incurs
> allocation overhead by distributing pages across pools, wasting memory
> on systems with fewer CPUs and less RAM.
>
> This patch allocates approximately 2*CPU zpools, with a minimum of 1
> zpool for single-CPU systems and up to 32 zpools for systems with 16 or
> more CPUs.  This number is sufficient to keep the probability of
> busy-waiting by a thread under 40%. The upper limit of 32 zpools remains
> unchanged.
>
> For memory, it limits to 1 zpool per 60MB of memory for the 20% default
> max pool size limit, assuming the best case with no fragmentation in
> zspages. It expects 90% pool usage for zsmalloc.
>
> Signed-off-by: Takero Funaki <flintglass@...il.com>
> ---

I think this needs benchmarking. Theoretical justification is nice,
but I'm not convinced it'll translate neatly to the messy world of
real life systems.