Re: [percpu] ace7e70901: aim9.sync_disk_rw.ops_per_sec -2.3% regression

[Dennis Zhou]

On Tue, May 11, 2021 at 10:26:14AM +0800, Oliver Sang wrote:
> Hi Dennis,
> 
> On Fri, May 07, 2021 at 07:08:03PM +0000, Dennis Zhou wrote:
> > On Fri, May 07, 2021 at 10:52:22AM -0700, Roman Gushchin wrote:
> > > On Fri, May 07, 2021 at 11:06:06AM +0800, Oliver Sang wrote:
> > > > hi Roman,
> > > >  
> > > > On Thu, May 06, 2021 at 12:54:59AM +0000, Roman Gushchin wrote:
> > > > > Ping
> > > > 
> > > > sorry for late.
> > > > 
> > > > the new patch makes the performance a little better but still has
> > > > 1.9% regression comparing to
> > > > f183324133 ("percpu: implement partial chunk depopulation")
> > > 
> > > Hi Oliver!
> > > 
> > > Thank you for testing it!
> > > 
> > > Btw, can you, please, confirm that the regression is coming specifically
> > > from ace7e70901 ("percpu: use reclaim threshold instead of running for every page")?
> > > I do see *some* regression in my setup, but the data is very noisy, so I'm not sure
> > > I can confirm it.
> > > 
> > > Thanks!
> > 
> > Thanks Oliver and Roman. If this is the case, I'll drop the final patch
> > and just merge up to f183324133 ("percpu: implement partial chunk
> > depopulation") into for-next as this is v5.14 anyway.
> > 
> > Oliver, is there a way to trigger the kernel test robot for a specific
> > test?
> 
> sorry for late.

No worries. Thanks for all you work!

> not sure what kind of specific test you want robot to do?
> if you mean for-next branch, if the branch is monitored by kernel test robot,
> after merge, it will be tested by robot automatically and the bisect will be
> triggered if there is still regression.

In this case, we believe there is a regression in
"aim9.sync_disk_rw.ops_per_sec". I know my branches are monitored (hence
we suspect this regression), but it would be nice to be able to kick off
a test with a patch or set of patches on top to validate that the
regression is fixed on your hardware configuration. Unfortunately I
don't have a 100+ core machine lying around :P.

Sorry for the additional questions, but is there a time frame that the
kernel robot is expected scrape over my tree / what test suites get run
against any particular branch?

> I found the ace7e70901 has already been dropped from original branch (dennis-percpu/for-5.14),

Yeah I have temporarily dropped it to get the others into for-next for
now. I'll spend some time later this week digging deeper into this.

> and we have data for this branch as below. from data, the f183324133 (current
> branch tip) doesn't introduce regression comparing 5.12-rc7 in our tests.
> 
> f183324133ea5 percpu: implement partial chunk depopulation                 103673.09 102188.39 104325.06 104038.4 102908.57 104057.06
> 1c29a3ceaf5f0 percpu: use pcpu_free_slot instead of pcpu_nr_slots - 1      104777.31 102225.93 101657.6
> 8ea2e1e35d1eb percpu: factor out pcpu_check_block_hint()                   102290.78 101853.87 102541.65
> d434405aaab7d Linux 5.12-rc7                                               102103.06 102248.12 101906.81 103033.13 102043.33
> 

Thanks,
Dennis