On Mon, Jun 27, 2022 at 8:25 AM Feng Tang [off-list ref] wrote:

On Mon, Jun 27, 2022 at 07:52:55AM -0700, Shakeel Butt wrote:

quoted

On Mon, Jun 27, 2022 at 5:34 AM Feng Tang [off-list ref] wrote:

quoted

Yes, 1% is just around noise level for a microbenchmark.

I went check the original test data of Oliver's report, the tests was
run 6 rounds and the performance data is pretty stable (0Day's report
will show any std deviation bigger than 2%)

The test platform is a 4 sockets 72C/144T machine, and I run the
same job (nr_tasks = 25% * nr_cpus) on one CascadeLake AP (4 nodes)
and one Icelake 2 sockets platform, and saw 75% and 53% regresson on
them.

In the first email, there is a file named 'reproduce', it shows the
basic test process:

"
  use 'performane' cpufre  governor for all CPUs

  netserver -4 -D
  modprobe sctp
  netperf -4 -H 127.0.0.1 -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K  &
  netperf -4 -H 127.0.0.1 -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K  &
  netperf -4 -H 127.0.0.1 -t SCTP_STREAM_MANY -c -C -l 300 -- -m 10K  &
  (repeat 36 times in total)
  ...

"

Which starts 36 (25% of nr_cpus) netperf clients. And the clients number
also matters, I tried to increase the client number from 36 to 72(50%),
and the regression is changed from 69.4% to 73.7%

Am I understanding correctly that this 69.4% (or 73.7%) regression is
with cgroup v2?

Yes.

quoted

Eric did the experiments on v2 but on real hardware where the
performance impact was negligible.

BTW do you see similar regression for tcp as well or just sctp?

Yes, I run TCP_SENDFILE case with 'send_size'==10K, it hits a
70%+ regressioin.

Thanks Feng. I think we should start with squeezing whatever we can
from layout changes and then try other approaches like increasing
batch size or something else. I can take a stab at this next week.

thanks,
Shakeel