Re: [PATCH] ipc/msg.c: mitigate the lock contention with percpu counter

[Sun, Jiebin]

On 9/3/2022 12:27 AM, Shakeel Butt wrote:
> On Fri, Sep 2, 2022 at 12:04 AM Jiebin Sun <jiebin.sun@xxxxxxxxx> wrote:
> > The msg_bytes and msg_hdrs atomic counters are frequently
> > updated when IPC msg queue is in heavy use, causing heavy
> > cache bounce and overhead. Change them to percpu_counters
> > greatly improve the performance. Since there is one unique
> > ipc namespace, additional memory cost is minimal. Reading
> > of the count done in msgctl call, which is infrequent. So
> > the need to sum up the counts in each CPU is infrequent.
> >
> > Apply the patch and test the pts/stress-ng-1.4.0
> > -- system v message passing (160 threads).
> >
> > Score gain: 3.38x
> >
> > CPU: ICX 8380 x 2 sockets
> > Core number: 40 x 2 physical cores
> > Benchmark: pts/stress-ng-1.4.0
> > -- system v message passing (160 threads)
> >
> > Signed-off-by: Jiebin Sun <jiebin.sun@xxxxxxxxx>
> [...]
> > +void percpu_counter_add_local(struct percpu_counter *fbc, s64 amount)
> > +{
> > +       this_cpu_add(*fbc->counters, amount);
> > +}
> > +EXPORT_SYMBOL(percpu_counter_add_local);
> Why not percpu_counter_add()? This may drift the fbc->count more than
> batch*nr_cpus. I am assuming that is not the issue for you as you
> always do an expensive sum in the slow path. As Andrew asked, this
> should be a separate patch.

Yes. It will always do sum in msgctl_info. So there is no need to
do global updating in percpu_counter_add when the percpu counter
reaches the batch size. We add percpu_counter_add_local in this
case. The sum in slow path is infrequent. So the additional cost
is much less compared to the atomic updating in do_msgsnd and
do_msgrcv every time. I have separate the original patch into two
patches.

Thanks.