[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <62bf4a21-ae1a-4ed0-8b55-9b444d9ca0a7@linux.dev>
Date: Tue, 18 Jun 2024 00:59:54 -0700
From: Chengming Zhou <chengming.zhou@...ux.dev>
To: Yu Zhao <yuzhao@...gle.com>
Cc: Minchan Kim <minchan@...nel.org>,
Sergey Senozhatsky <senozhatsky@...omium.org>,
Andrew Morton <akpm@...ux-foundation.org>,
Johannes Weiner <hannes@...xchg.org>, Yosry Ahmed <yosryahmed@...gle.com>,
Nhat Pham <nphamcs@...il.com>, Takero Funaki <flintglass@...il.com>,
Chengming Zhou <zhouchengming@...edance.com>, linux-mm@...ck.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/2] mm/zswap: use only one pool in zswap
On 2024/6/17 13:16, Yu Zhao wrote:
> On Mon, Jun 17, 2024 at 6:58 AM Chengming Zhou <chengming.zhou@...ux.dev> wrote:
>>
>> Zswap uses 32 pools to workaround the locking scalability problem in
>> zsmalloc,
>
> Note that zpool can have other backends (zbud, z3fold), and the
> original patch was developed (even before zswap could use zsmalloc) to
> make sure it works for all the backend.
>
> This patch only makes sense now only because zsmalloc became a lot
> more popular than other backends (even though some distros still
> default to zbud).
>
Right, we mostly focus on zsmalloc, I just do the same testing using zbud:
real user sys
6.10.0-rc3-zbud 138.23 1239.58 1430.09
6.10.0-rc3-onepool-zbud 139.64 1241.37 1516.59
Since we don't do any locking optimization for zbud, so performance
is worse than 32 pools as expected.
Actually Yosry did a great testing [1] to compare zsmalloc, zbud and
z3flod, to support that we can just use zsmalloc in zswap, so we can
drop zpool, zbud and z3flod.
I will include this information in the changelog in the next version.
[1]
https://lore.kernel.org/lkml/CAJD7tkbRF6od-2x_L8-A1QL3=2Ww13sCj4S3i4bNndqF+3+_Vg@mail.gmail.com/
Thanks.
Powered by blists - more mailing lists