Re: [fsnotify] 4c40d6efc8: unixbench.score -3.3% regression

From: Amir Goldstein <amir73il@gmail.com>
To: Oliver Sang <oliver.sang@intel.com>
Cc: Gabriel Krisman Bertazi <krisman@collabora.com>,
	LKML <linux-kernel@vger.kernel.org>,
	lkp@lists.01.org, ying.huang@intel.com, feng.tang@intel.com,
	zhengjun.xing@linux.intel.com, Jan Kara <jack@suse.com>,
	"Darrick J. Wong" <djwong@kernel.org>,
	Theodore Tso <tytso@mit.edu>, Dave Chinner <david@fromorbit.com>,
	David Howells <dhowells@redhat.com>,
	Khazhismel Kumykov <khazhy@google.com>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Ext4 <linux-ext4@vger.kernel.org>,
	kernel@collabora.com, Mel Gorman <mgorman@techsingularity.net>,
	0day robot <lkp@intel.com>
Subject: Re: [fsnotify] 4c40d6efc8: unixbench.score -3.3% regression
Date: Tue, 3 Aug 2021 19:19:04 +0300	[thread overview]
Message-ID: <CAOQ4uxisyDjVpWX1M6O4ugxBbcX+LWWf4NQJ+LQY1-3-9tN+BA@mail.gmail.com> (raw)
In-Reply-To: <20210803142225.GA28609@xsang-OptiPlex-9020>

> > Oliver,
> >
> > Would it be possible to request a re-test with the branch:
> > https://github.com/amir73il/linux fsnotify-perf
> >
> > The patch at the tip of that branch is the one this regression report
> > has blamed.
> >
> > My expectation is that the patch at fsnotify-perf^ ("fsnotify: optimize the
> > case of no marks of any type") will improve performance of the test case
> > compared to baseline (v5.14-rc3) and that the patch at the tip of fsnotify-perf
> > would not regress performance.
>
> we tested this branch and the results meet your expectation.
>
> fsnotify-perf^ improves performance comparing to v5.14-rc3. tip is a little worse
> than its parent (-3.3%), but still better than v5.14-rc3.
>
> below is detail data.
>
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/nr_task/rootfs/runtime/tbox_group/test/testcase/ucode:
>   gcc-9/performance/x86_64-rhel-8.3/1/debian-10.4-x86_64-20200603.cgz/300s/lkp-csl-2sp4/pipe/unixbench/0x4003006
>
> commit:
>   v5.14-rc3
>   23050d041 ("fsnotify: optimize the case of no marks of any type")
>   7446ba772 ("fsnotify: pass arguments of fsnotify() in struct fsnotify_event_info")
>
>        v5.14-rc3 23050d0419441a02185e4ed5170 7446ba772ae107ab937cd04e880
> ---------------- --------------------------- ---------------------------
>          %stddev     %change         %stddev     %change         %stddev
>              \          |                \          |                \
>       1562            +8.0%       1688            +4.5%       1633        unixbench.score

Hi Oliver,

Thanks a lot for testing!

I don't know what to make of the (-3.3%) degradation because I was expecting
that fsnotify-perf^ would optimize out the calls to fsnotify() and fsnotify-perf
only changes code from fsnotify() and below.

But I guess it doesn't matter much as Gabriel said, its a cleanup patch and
we can drop it.

But now that I have this report I can post the fsnotify-perf^ patches :-)

Thanks,
Amir.