Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220619150456.GB34471@xsang-OptiPlex-9020> <20220622172857.37db0d29@kernel.org>
 <CADvbK_csvmkKe46hT9792=+Qcjor2EvkkAnr--CJK3NGX-N9BQ@mail.gmail.com>
 <CADvbK_eQUmb942vC+bG+NRzM1ki1LiCydEDR1AezZ35Jvsdfnw@mail.gmail.com>
 <20220623185730.25b88096@kernel.org> <CANn89iLidqjiiV8vxr7KnUg0JvfoS9+TRGg=8ANZ8NBRjeQxsQ@mail.gmail.com>
In-Reply-To: <CANn89iLidqjiiV8vxr7KnUg0JvfoS9+TRGg=8ANZ8NBRjeQxsQ@mail.gmail.com>
From:   Eric Dumazet <edumazet@google.com>
Date:   Fri, 24 Jun 2022 06:22:42 +0200
Message-ID: <CANn89iLZYGOjTFThvq-J678BtPtqsywwxFekt9XTcPxk_La8eg@mail.gmail.com>
Subject: Re: [net] 4890b686f4: netperf.Throughput_Mbps -69.4% regression
To:     Jakub Kicinski <kuba@kernel.org>
Cc:     Xin Long <lucien.xin@gmail.com>,
        Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>,
        kernel test robot <oliver.sang@intel.com>,
        Shakeel Butt <shakeelb@google.com>,
        Soheil Hassas Yeganeh <soheil@google.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Linux Memory Management List <linux-mm@kvack.org>,
        network dev <netdev@vger.kernel.org>,
        linux-s390@vger.kernel.org, MPTCP Upstream <mptcp@lists.linux.dev>,
        "linux-sctp @ vger . kernel . org" <linux-sctp@vger.kernel.org>,
        lkp@lists.01.org, kbuild test robot <lkp@intel.com>,
        Huang Ying <ying.huang@intel.com>,
        "Tang, Feng" <feng.tang@intel.com>, zhengjun.xing@linux.intel.com,
        fengwei.yin@intel.com, Ying Xu <yinxu@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Jun 24, 2022 at 6:13 AM Eric Dumazet <edumazet@google.com> wrote:
>
> On Fri, Jun 24, 2022 at 3:57 AM Jakub Kicinski <kuba@kernel.org> wrote:
> >
> > On Thu, 23 Jun 2022 18:50:07 -0400 Xin Long wrote:
> > > From the perf data, we can see __sk_mem_reduce_allocated() is the one
> > > using CPU the most more than before, and mem_cgroup APIs are also
> > > called in this function. It means the mem cgroup must be enabled in
> > > the test env, which may explain why I couldn't reproduce it.
> > >
> > > The Commit 4890b686f4 ("net: keep sk->sk_forward_alloc as small as
> > > possible") uses sk_mem_reclaim(checking reclaimable >= PAGE_SIZE) to
> > > reclaim the memory, which is *more frequent* to call
> > > __sk_mem_reduce_allocated() than before (checking reclaimable >=
> > > SK_RECLAIM_THRESHOLD). It might be cheap when
> > > mem_cgroup_sockets_enabled is false, but I'm not sure if it's still
> > > cheap when mem_cgroup_sockets_enabled is true.
> > >
> > > I think SCTP netperf could trigger this, as the CPU is the bottleneck
> > > for SCTP netperf testing, which is more sensitive to the extra
> > > function calls than TCP.
> > >
> > > Can we re-run this testing without mem cgroup enabled?
> >
> > FWIW I defer to Eric, thanks a lot for double checking the report
> > and digging in!
>
> I did tests with TCP + memcg and noticed a very small additional cost
> in memcg functions,
> because of suboptimal layout:
>
> Extract of an internal Google bug, update from June 9th:
>
> --------------------------------
> I have noticed a minor false sharing to fetch (struct
> mem_cgroup)->css.parent, at offset 0xc0,
> because it shares the cache line containing struct mem_cgroup.memory,
> at offset 0xd0
>
> Ideally, memcg->socket_pressure and memcg->parent should sit in a read
> mostly cache line.
> -----------------------
>
> But nothing that could explain a "-69.4% regression"

I guess the test now hits memcg limits more often, forcing expensive reclaim,
and the memcg limits need some adjustments.

Overall, tests enabling memcg should probably need fine tuning, I will
defer to Intel folks.


>
> memcg has a very similar strategy of per-cpu reserves, with
> MEMCG_CHARGE_BATCH being 32 pages per cpu.
>
> It is not clear why SCTP with 10K writes would overflow this reserve constantly.
>
> Presumably memcg experts will have to rework structure alignments to
> make sure they can cope better
> with more charge/uncharge operations, because we are not going back to
> gigantic per-socket reserves,
> this simply does not scale.