On Mon, Jun 27, 2022 at 4:53 PM Shakeel Butt [off-list ref] wrote:

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

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?

TCP_RR with big packets can show a regression as well.

I gave this perf profile:

    28.69%  [kernel]       [k] copy_user_enhanced_fast_string
    16.13%  [kernel]       [k] intel_idle_irq
     6.46%  [kernel]       [k] page_counter_try_charge
     6.20%  [kernel]       [k] __sk_mem_reduce_allocated
     5.68%  [kernel]       [k] try_charge_memcg
     5.16%  [kernel]       [k] page_counter_cancel

And this points to false sharing on (struct page_counter *)->usage

I guess memcg had free lunch, because of per-socket cache, that we
need to remove.