Re: [RFC PATCH v0] mm/slub: Let number of online CPUs determine the slub page order

[Vincent Guittot]

On Tue, 26 Jan 2021 at 09:52, Michal Hocko <mhocko@xxxxxxxx> wrote:
>
> On Thu 21-01-21 19:19:21, Vlastimil Babka wrote:
> [...]
> > We could also start questioning the very assumption that number of cpus should
> > affect slab page size in the first place. Should it? After all, each CPU will
> > have one or more slab pages privately cached, as we discuss in the other
> > thread... So why make the slab pages also larger?
>
> I do agree. What is the acutal justification for this scaling?
>         /*
>          * Attempt to find best configuration for a slab. This
>          * works by first attempting to generate a layout with
>          * the best configuration and backing off gradually.
>          *
>          * First we increase the acceptable waste in a slab. Then
>          * we reduce the minimum objects required in a slab.
>          */
>
> doesn't speak about CPUs.  9b2cd506e5f2 ("slub: Calculate min_objects
> based on number of processors.") does talk about hackbench "This has
> been shown to address the performance issues in hackbench on 16p etc."
> but it doesn't give any more details to tell actually _why_ that works.
>
> This thread shows that this is still somehow related to performance but
> the real reason is not clear. I believe we should be focusing on the
> actual reasons for the performance impact than playing with some fancy
> math and tuning for a benchmark on a particular machine which doesn't
> work for others due to subtle initialization timing issues.
>
> Fundamentally why should higher number of CPUs imply the size of slab in
> the first place?

A 1st answer is that the activity and the number of threads involved
scales with the number of CPUs. Regarding the hackbench benchmark as
an example, the number of group/threads raise to a higher level on the
server than on the small system which doesn't seem unreasonable.

On 8 CPUs, I run hackbench with up to 16 groups which means 16*40
threads. But I raise up to 256 groups, which means 256*40 threads, on
the 224 CPUs system. In fact, hackbench -g 1 (with 1 group) doesn't
regress on the 224 CPUs  system.  The next test with 4 groups starts
to regress by -7%. But the next one: hackbench -g 16 regresses by 187%
(duration is almost 3 times longer). It seems reasonable to assume
that the number of running threads and resources scale with the number
of CPUs because we want to run more stuff.


> --
> Michal Hocko
> SUSE Labs