Re: [amir73il:fanotify_pre_content] [fanotify] 2731bef965: fxmark.ssd_xfs_MRPL_72_bufferedio.works/sec -12.0% regression

Re: [amir73il:fanotify_pre_content] [fanotify] 2731bef965: fxmark.ssd_xfs_MRPL_72_bufferedio.works/sec -12.0% regression

[Yujie Liu]

On Wed, Oct 26, 2022 at 06:06:45PM +0300, Amir Goldstein wrote:
> On Mon, Oct 24, 2022 at 4:34 AM kernel test robot <yujie.liu@intel.com> wrote:
> >
> > Greeting,
> >
> > FYI, we noticed a -12.0% regression of fxmark.ssd_xfs_MRPL_72_bufferedio.works/sec due to commit:
> >
> > commit: 2731bef9653952cab508c5eb701592ea78dc6e75 ("fanotify: introduce FAN_LOOKUP_PERM event")
> > https://github.com/amir73il/linux fanotify_pre_content
> >
> > in testcase: fxmark
> > on test machine: 128 threads 2 sockets Intel(R) Xeon(R) Platinum 8358 CPU @ 2.60GHz (Ice Lake) with 128G memory
> > with following parameters:
> >
> >         disk: 1SSD
> >         media: ssd
> >         test: MRPL
> >         fstype: xfs
> >         directio: bufferedio
> >         cpufreq_governor: performance
> >
> > test-description: FxMark is a filesystem benchmark that test multicore scalability.
> > test-url: https://github.com/sslab-gatech/fxmark
> >
> >
> > If you fix the issue, kindly add following tag
> > | Reported-by: kernel test robot <yujie.liu@intel.com>
> >
> >
> > Details are as below:
> >
> > =========================================================================================
> > compiler/cpufreq_governor/directio/disk/fstype/kconfig/media/rootfs/tbox_group/test/testcase:
> >   gcc-11/performance/bufferedio/1SSD/xfs/x86_64-rhel-8.3/ssd/debian-11.1-x86_64-20220510.cgz/lkp-icl-2sp5/MRPL/fxmark
> >
> > commit:
> >   da8d464cb0 ("fanotify: allow O_PATH flag in event_f_flags")
> >   2731bef965 ("fanotify: introduce FAN_LOOKUP_PERM event")
> 
> Hello!
> 
> Thanks for the report.
> 
> Could you please run this test on branch
> https://github.com/amir73il/fan_lookup_perm
> 
>   2731bef96539 ("fanotify: introduce FAN_LOOKUP_PERM event")
>   87f9aa515fd0 ("fsnotify: optimize fsnotify_lookup_perm() hook")
> 
> To see if it fixes the reported regression?

Hi Amir,

Thanks for the fix, we tested above commits and found that the
regression has been resolved.

  Tested-by: Yujie Liu <yujie.liu@intel.com>

=========================================================================================
compiler/cpufreq_governor/directio/disk/fstype/kconfig/media/rootfs/tbox_group/test/testcase:
  gcc-11/performance/bufferedio/1SSD/xfs/x86_64-rhel-8.3/ssd/debian-11.1-x86_64-20220510.cgz/lkp-icl-2sp5/MRPL/fxmark

commit:
  da8d464cb0251 ("fanotify: allow O_PATH flag in event_f_flags")
  2731bef965395 ("fanotify: introduce FAN_LOOKUP_PERM event")
  87f9aa515fd01 ("fsnotify: optimize fsnotify_lookup_perm() hook")

da8d464cb02517d3 2731bef9653952cab508c5eb701 87f9aa515fd01eabda2b1ac2707
---------------- --------------------------- ---------------------------
         %stddev     %change         %stddev     %change         %stddev
             \          |                \          |                \
     44.04           -11.7%      38.90            -1.0%      43.62        fxmark.ssd_xfs_MRPL_18_bufferedio.user_sec
      4.89           -11.7%       4.32            -1.0%       4.84        fxmark.ssd_xfs_MRPL_18_bufferedio.user_util
 1.624e+09           -12.0%   1.43e+09            -0.7%  1.613e+09        fxmark.ssd_xfs_MRPL_18_bufferedio.works
  32484848           -12.0%   28590324            -0.7%   32266981        fxmark.ssd_xfs_MRPL_18_bufferedio.works/sec
      2.51 ±  4%     -12.1%       2.20 ±  3%      +0.7%       2.52 ±  2%  fxmark.ssd_xfs_MRPL_1_bufferedio.user_sec
      5.01 ±  4%     -12.1%       4.41 ±  3%      +0.7%       5.05 ±  2%  fxmark.ssd_xfs_MRPL_1_bufferedio.user_util
  91798396           -12.1%   80692995            -0.4%   91437550        fxmark.ssd_xfs_MRPL_1_bufferedio.works
   1835968           -12.1%    1613860            -0.4%    1828751        fxmark.ssd_xfs_MRPL_1_bufferedio.works/sec
 1.847e+08           -11.9%  1.627e+08            -0.6%  1.837e+08        fxmark.ssd_xfs_MRPL_2_bufferedio.works
   3694940           -11.9%    3254103            -0.6%    3673299        fxmark.ssd_xfs_MRPL_2_bufferedio.works/sec
     88.28           -11.7%      77.95            -1.0%      87.38        fxmark.ssd_xfs_MRPL_36_bufferedio.user_sec
      4.90           -11.7%       4.33            -1.0%       4.85        fxmark.ssd_xfs_MRPL_36_bufferedio.user_util
 3.055e+09           -11.9%  2.692e+09            -0.9%  3.027e+09        fxmark.ssd_xfs_MRPL_36_bufferedio.works
  61109619           -11.9%   53847415            -0.9%   60548779        fxmark.ssd_xfs_MRPL_36_bufferedio.works/sec
     10.02 ±  2%     -12.2%       8.79            -0.9%       9.93 ±  2%  fxmark.ssd_xfs_MRPL_4_bufferedio.user_sec
      5.01 ±  2%     -12.2%       4.39            -0.9%       4.96 ±  2%  fxmark.ssd_xfs_MRPL_4_bufferedio.user_util
 3.699e+08           -11.8%  3.263e+08            -0.3%  3.689e+08        fxmark.ssd_xfs_MRPL_4_bufferedio.works
   7397404           -11.8%    6525259            -0.3%    7378874        fxmark.ssd_xfs_MRPL_4_bufferedio.works/sec
    132.91 ±  2%     -12.3%     116.60            -0.8%     131.91        fxmark.ssd_xfs_MRPL_54_bufferedio.user_sec
      4.92 ±  2%     -12.3%       4.32            -0.8%       4.88        fxmark.ssd_xfs_MRPL_54_bufferedio.user_util
 4.664e+09           -12.0%  4.105e+09            -0.7%  4.629e+09        fxmark.ssd_xfs_MRPL_54_bufferedio.works
  93279586           -12.0%   82101317            -0.7%   92583640        fxmark.ssd_xfs_MRPL_54_bufferedio.works/sec
    167.55           -12.7%     146.34            -0.7%     166.45        fxmark.ssd_xfs_MRPL_72_bufferedio.user_sec
      4.65           -12.7%       4.06            -0.7%       4.62        fxmark.ssd_xfs_MRPL_72_bufferedio.user_util
   5.6e+09           -12.0%  4.926e+09            -1.1%  5.539e+09        fxmark.ssd_xfs_MRPL_72_bufferedio.works
  1.12e+08           -12.0%   98517557            -1.1%  1.108e+08        fxmark.ssd_xfs_MRPL_72_bufferedio.works/sec


--
Best Regards,
Yujie