[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <7302d357-b42f-4874-af59-43ad03a0b47f@app.fastmail.com>
Date: Thu, 04 Sep 2025 13:07:44 -0400
From: "Chris Murphy" <lists@...orremedies.com>
To: "David Sterba" <dsterba@...e.cz>, "Qu WenRuo" <wqu@...e.com>
Cc: "kernel test robot" <oliver.sang@...el.com>, oe-lkp@...ts.linux.dev,
lkp@...el.com, linux-kernel <linux-kernel@...r.kernel.org>,
"David Sterba" <dsterba@...e.com>,
"Btrfs BTRFS" <linux-btrfs@...r.kernel.org>
Subject: Re: [linus:master] [btrfs] bddf57a707: stress-ng.sync-file.ops_per_sec 44.2%
regression
On Thu, Sep 4, 2025, at 10:34 AM, David Sterba wrote:
> On Wed, Sep 03, 2025 at 06:18:01PM +0930, Qu Wenruo wrote:
>> This doesn't sound sane to me.
>>
>> The two commits are only affecting btrfs mounting/unmounting, I can not
>> make any sense on why they would affect performance.
>>
>> Or does stress-ng doing a lot of mounting/unmounting?
>
> Yeah, unless there's some indirect way how mount affects the tests the
> numbers do not match the identified patches. The difference is roughly
> consistent in all the stats to be about 40% less so it's like it's doing
> half of the work. Delayed device opening does not explain that.
Does the test run on qemu/kvm? Could cache mode and host workload affect the result?
If it were unsafe mode, the guest very quickly thinks the write is on stable media even though the host can significantly delay writing to stable media. Whereas directsync mode might be really slow since the host must commit to stable media before the guest sees it as on stable media.
--
Chris Murphy
Powered by blists - more mailing lists