Re: [linux-next:master] [memcg] 70a64b7919: will-it-scale.per_process_ops -11.9% regression

[Shakeel Butt <shakeel.butt@linux.dev>]

On Mon, May 20, 2024 at 10:43:35AM +0800, Oliver Sang wrote:
> hi, Shakeel,
> 
> On Sun, May 19, 2024 at 10:20:28AM -0700, Shakeel Butt wrote:
> > On Sun, May 19, 2024 at 05:14:39PM +0800, Oliver Sang wrote:
> > > hi, Shakeel,
> > > 
> > > On Fri, May 17, 2024 at 11:28:10PM -0700, Shakeel Butt wrote:
> > > > On Fri, May 17, 2024 at 01:56:30PM +0800, kernel test robot wrote:
> > > > > 
> > > > > 
> > > > > Hello,
> > > > > 
> > > > > kernel test robot noticed a -11.9% regression of will-it-scale.per_process_ops on:
> > > > > 
> > > > > 
> > > > > commit: 70a64b7919cbd6c12306051ff2825839a9d65605 ("memcg: dynamically allocate lruvec_stats")
> > > > > https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
> > > > > 
> > > > 
> > > > Thanks for the report. Can you please run the same benchmark but with
> > > > the full series (of 8 patches) or at least include the ff48c71c26aa
> > > > ("memcg: reduce memory for the lruvec and memcg stats").
> > > 
> > > while this bisect, ff48c71c26aa has been checked. it has silimar data as
> > > 70a64b7919 (a little worse actually)
> > > 
> > > 59142d87ab03b8ff 70a64b7919cbd6c12306051ff28 ff48c71c26aaefb090c108d8803
> > > ---------------- --------------------------- ---------------------------
> > >          %stddev     %change         %stddev     %change         %stddev
> > >              \          |                \          |                \
> > >      91713           -11.9%      80789           -13.2%      79612        will-it-scale.per_process_ops
> > > 
> > > 
> > > ok, we will run tests on tip of the series which should be below if I understand
> > > it correctly.
> > > 
> > > * a94032b35e5f9 memcg: use proper type for mod_memcg_state
> > > 
> > > 
> > 
> > Thanks a lot Oliver. One question: what is the filesystem mounted at
> > /tmp on your test machine? I just wanted to make sure I run the test
> > with minimal changes from your setup.
> 
> we don't have specific partition for /tmp, just use tmpfs
> 
> tmp on /tmp type tmpfs (rw,relatime)
> 
> 
> BTW, the test on a94032b35e5f9 finished, still have similar score to 70a64b7919
> 
> =========================================================================================
> compiler/cpufreq_governor/kconfig/mode/nr_task/rootfs/tbox_group/test/testcase:
>   gcc-13/performance/x86_64-rhel-8.3/process/100%/debian-12-x86_64-20240206.cgz/lkp-skl-fpga01/page_fault2/will-it-scale
> 
> 59142d87ab03b8ff 70a64b7919cbd6c12306051ff28 ff48c71c26aaefb090c108d8803 a94032b35e5f97dc1023030d929
> ---------------- --------------------------- --------------------------- ---------------------------
>          %stddev     %change         %stddev     %change         %stddev     %change         %stddev
>              \          |                \          |                \          |                \
>      91713           -11.9%      80789           -13.2%      79612           -13.0%      79833        will-it-scale.per_process_ops
> 

Thanks again. I am not sure if you have a single node machine but if you
have, can you try to repro this issue on such machine. At the moment, I
don't have access to such machine but I will try to repro myself as
well.

Shakeel