From: Roman Gushchin <guro@fb.com>
To: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>,
Shakeel Butt <shakeelb@google.com>,
Andrew Morton <akpm@linux-foundation.org>,
Christoph Lameter <cl@linux.com>,
Johannes Weiner <hannes@cmpxchg.org>,
Michal Hocko <mhocko@kernel.org>, Linux MM <linux-mm@kvack.org>,
Kernel Team <kernel-team@fb.com>,
LKML <linux-kernel@vger.kernel.org>,
Mel Gorman <mgorman@techsingularity.net>
Subject: Re: [PATCH v6 00/19] The new cgroup slab memory controller
Date: Thu, 18 Jun 2020 18:30:13 -0700 [thread overview]
Message-ID: <20200619013013.GE135965@carbon.dhcp.thefacebook.com> (raw)
In-Reply-To: <20200618113121.78a6a2ca@carbon>
On Thu, Jun 18, 2020 at 11:31:21AM +0200, Jesper Dangaard Brouer wrote:
> On Thu, 18 Jun 2020 10:43:44 +0200
> Jesper Dangaard Brouer <brouer@redhat.com> wrote:
>
> > On Wed, 17 Jun 2020 18:29:28 -0700
> > Roman Gushchin <guro@fb.com> wrote:
> >
> > > On Wed, Jun 17, 2020 at 01:24:21PM +0200, Vlastimil Babka wrote:
> > > > On 6/17/20 5:32 AM, Roman Gushchin wrote:
> > > > > On Tue, Jun 16, 2020 at 08:05:39PM -0700, Shakeel Butt wrote:
> > > > >> On Tue, Jun 16, 2020 at 7:41 PM Roman Gushchin <guro@fb.com> wrote:
> > > > >> >
> > > > >> > On Tue, Jun 16, 2020 at 06:46:56PM -0700, Shakeel Butt wrote:
> > > > >> > > On Mon, Jun 8, 2020 at 4:07 PM Roman Gushchin <guro@fb.com> wrote:
> > > > >> > > >
> > > > >> [...]
> > > > >> > >
> > > > >> > > Have you performed any [perf] testing on SLAB with this patchset?
> > > > >> >
> > > > >> > The accounting part is the same for SLAB and SLUB, so there should be no
> > > > >> > significant difference. I've checked that it compiles, boots and passes
> > > > >> > kselftests. And that memory savings are there.
> > > > >> >
> > > > >>
> > > > >> What about performance? Also you mentioned that sharing kmem-cache
> > > > >> between accounted and non-accounted can have additional overhead. Any
> > > > >> difference between SLAB and SLUB for such a case?
> > > > >
> > > > > Not really.
> > > > >
> > > > > Sharing a single set of caches adds some overhead to root- and non-accounted
> > > > > allocations, which is something I've tried hard to avoid in my original version.
> > > > > But I have to admit, it allows to simplify and remove a lot of code, and here
> > > > > it's hard to argue with Johanness, who pushed on this design.
> > > > >
> > > > > With performance testing it's not that easy, because it's not obvious what
> > > > > we wanna test. Obviously, per-object accounting is more expensive, and
> > > > > measuring something like 1000000 allocations and deallocations in a line from
> > > > > a single kmem_cache will show a regression. But in the real world the relative
> > > > > cost of allocations is usually low, and we can get some benefits from a smaller
> > > > > working set and from having shared kmem_cache objects cache hot.
> > > > > Not speaking about some extra memory and the fragmentation reduction.
> > > > >
> > > > > We've done an extensive testing of the original version in Facebook production,
> > > > > and we haven't noticed any regressions so far. But I have to admit, we were
> > > > > using an original version with two sets of kmem_caches.
> > > > >
> > > > > If you have any specific tests in mind, I can definitely run them. Or if you
> > > > > can help with the performance evaluation, I'll appreciate it a lot.
> > > >
> > > > Jesper provided some pointers here [1], it would be really great if you could
> > > > run at least those microbenchmarks. With mmtests it's the major question of
> > > > which subset/profiles to run, maybe the referenced commits provide some hints,
> > > > or maybe Mel could suggest what he used to evaluate SLAB vs SLUB not so long ago.
> > > >
> > > > [1] https://lore.kernel.org/linux-mm/20200527103545.4348ac10@carbon/
> > >
> > > Oh, Jesper, I'm really sorry, somehow I missed your mail.
> > > Thank you, Vlastimil, for pointing at it.
> > >
> > > I've got some results (slab_bulk_test01), but honestly I fail to interpret them.
> > >
> > > I ran original vs patched with SLUB and SLAB, each test several times and picked
> > > 3 which looked most consistently. But it still looks very noisy.
> > >
> > > I ran them on my desktop (8-core Ryzen 1700, 16 GB RAM, Fedora 32),
> > > it's 5.8-rc1 + slab controller v6 vs 5.8-rc1 (default config from Fedora 32).
> >
> > What about running these tests on the server level hardware, that you
> > intent to run this on?
>
> To give you an idea of the performance difference I ran the same test
> on a Broadwell Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz.
>
> The SLUB fastpath:
> Type:kmem fastpath reuse Per elem: 60 cycles(tsc) 16.822 ns
>
>
> > >
> > > How should I interpret this data?
> >
> > First of all these SLUB+SLAB microbenchmarks use object size 256 bytes,
> > because network stack alloc object of this size for SKBs/sk_buff (due
> > to cache-align as used size is 224 bytes). Checked SLUB: Each slab use
> > 2 pages (8192 bytes) and contain 32 object of size 256 (256*32=8192).
> >
> > The SLUB allocator have a per-CPU slab which speedup fast-reuse, in this
> > case up-to 32 objects. For SLUB the "fastpath reuse" test this behaviour,
> > and it serves as a baseline for optimal 1-object performance (where my bulk
> > API tries to beat that, which is possible even for 1-object due to knowing
> > bulk API cannot be used from IRQ context).
> >
> > SLUB fastpath: 3 measurements reporting cycles(tsc)
> > - SLUB-patched : fastpath reuse: 184 - 177 - 176 cycles(tsc)
> > - SLUB-original: fastpath reuse: 178 - 153 - 156 cycles(tsc)
> >
>
> For your SLAB results:
>
> SLAB fastpath: 3 measurements reporting cycles(tsc)
> - SLAB-patched : 161 - 160 - 163 cycles(tsc)
> - SLAB-original: 174 - 170 - 159 cycles(tsc)
>
> I find it strange that SLAB is slightly better than SLUB (in many
> measurements), because SLUB should have an advantage on this fast-path
> quick reuse due to the per-CPU slabs. Maybe this is also related to
> the CPU arch you are using?
>
>
> > There are some stability concerns as you mention, but it seems pretty
> > consistently that patched version is slower. If you compile with
> > no-PREEMPT you can likely get more stable results (and remove a slight
> > overhead for SLUB fastpath).
> >
> > The microbenchmark also measures the bulk-API, which is AFAIK only used
> > by network stack (and io_uring). I guess you shouldn't focus too much
> > on these bulk measurements. When bulk-API cross this objects per slab
> > threshold, or is unlucky is it use two per-CPU slab, then the
> > measurements can fluctuate a bit.
> >
> > Your numbers for SLUB bulk-API:
> >
> > SLUB-patched - bulk-API
> > - SLUB-patched : bulk_quick_reuse objects=1 : 187 - 90 - 224 cycles(tsc)
> > - SLUB-patched : bulk_quick_reuse objects=2 : 110 - 53 - 133 cycles(tsc)
> > - SLUB-patched : bulk_quick_reuse objects=3 : 88 - 95 - 42 cycles(tsc)
> > - SLUB-patched : bulk_quick_reuse objects=4 : 91 - 85 - 36 cycles(tsc)
> > - SLUB-patched : bulk_quick_reuse objects=8 : 32 - 66 - 32 cycles(tsc)
> >
> > SLUB-original - bulk-API
> > - SLUB-original: bulk_quick_reuse objects=1 : 87 - 87 - 142 cycles(tsc)
> > - SLUB-original: bulk_quick_reuse objects=2 : 52 - 53 - 53 cycles(tsc)
> > - SLUB-original: bulk_quick_reuse objects=3 : 42 - 42 - 91 cycles(tsc)
> > - SLUB-original: bulk_quick_reuse objects=4 : 91 - 37 - 37 cycles(tsc)
> > - SLUB-original: bulk_quick_reuse objects=8 : 31 - 79 - 76 cycles(tsc)
>
> Your numbers for SLAB bulk-API:
>
> SLAB-patched - bulk-API
> - SLAB-patched : bulk_quick_reuse objects=1 : 67 - 67 - 140 cycles(tsc)
> - SLAB-patched : bulk_quick_reuse objects=2 : 55 - 46 - 46 cycles(tsc)
> - SLAB-patched : bulk_quick_reuse objects=3 : 93 - 94 - 39 cycles(tsc)
> - SLAB-patched : bulk_quick_reuse objects=4 : 35 - 88 - 85 cycles(tsc)
> - SLAB-patched : bulk_quick_reuse objects=8 : 30 - 30 - 30 cycles(tsc)
>
> SLAB-original- bulk-API
> - SLAB-original: bulk_quick_reuse objects=1 : 143 - 136 - 67 cycles(tsc)
> - SLAB-original: bulk_quick_reuse objects=2 : 45 - 46 - 46 cycles(tsc)
> - SLAB-original: bulk_quick_reuse objects=3 : 38 - 39 - 39 cycles(tsc)
> - SLAB-original: bulk_quick_reuse objects=4 : 35 - 87 - 87 cycles(tsc)
> - SLAB-original: bulk_quick_reuse objects=8 : 29 - 66 - 30 cycles(tsc)
>
> In case of SLAB I expect the bulk-API to be slightly faster than SLUB,
> as the SLUB bulk code is much more advanced.
So again it looks like a patched version is only slightly worse if we're taking
the smallest number in each series. Is it a correct assumption?
Thanks!
next prev parent reply other threads:[~2020-06-19 1:30 UTC|newest]
Thread overview: 92+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-06-08 23:06 [PATCH v6 00/19] The new cgroup slab memory controller Roman Gushchin
2020-06-08 23:06 ` [PATCH v6 01/19] mm: memcg: factor out memcg- and lruvec-level changes out of __mod_lruvec_state() Roman Gushchin
2020-06-17 1:52 ` Shakeel Butt
2020-06-17 2:50 ` Roman Gushchin
2020-06-17 2:59 ` Shakeel Butt
2020-06-17 3:19 ` Roman Gushchin
2020-06-08 23:06 ` [PATCH v6 02/19] mm: memcg: prepare for byte-sized vmstat items Roman Gushchin
2020-06-17 2:57 ` Shakeel Butt
2020-06-17 3:19 ` Roman Gushchin
2020-06-17 15:55 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 03/19] mm: memcg: convert vmstat slab counters to bytes Roman Gushchin
2020-06-17 3:03 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 04/19] mm: slub: implement SLUB version of obj_to_index() Roman Gushchin
2020-06-17 3:08 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 05/19] mm: memcontrol: decouple reference counting from page accounting Roman Gushchin
2020-06-18 0:47 ` Shakeel Butt
2020-06-18 14:55 ` Shakeel Butt
2020-06-18 19:51 ` Roman Gushchin
2020-06-19 1:08 ` Roman Gushchin
2020-06-19 1:18 ` Shakeel Butt
2020-06-19 1:31 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 06/19] mm: memcg/slab: obj_cgroup API Roman Gushchin
2020-06-19 15:42 ` Shakeel Butt
2020-06-19 21:38 ` Roman Gushchin
2020-06-19 22:16 ` Shakeel Butt
2020-06-19 22:52 ` Roman Gushchin
2020-06-20 22:50 ` Andrew Morton
2020-06-08 23:06 ` [PATCH v6 07/19] mm: memcg/slab: allocate obj_cgroups for non-root slab pages Roman Gushchin
2020-06-19 16:36 ` Shakeel Butt
2020-06-20 0:25 ` Roman Gushchin
2020-06-20 0:31 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 08/19] mm: memcg/slab: save obj_cgroup for non-root slab objects Roman Gushchin
2020-06-20 0:16 ` Shakeel Butt
2020-06-20 1:19 ` Roman Gushchin
2020-06-08 23:06 ` [PATCH v6 09/19] mm: memcg/slab: charge individual slab objects instead of pages Roman Gushchin
2020-06-20 0:54 ` Shakeel Butt
2020-06-20 1:29 ` Roman Gushchin
2020-06-08 23:06 ` [PATCH v6 10/19] mm: memcg/slab: deprecate memory.kmem.slabinfo Roman Gushchin
2020-06-22 17:12 ` Shakeel Butt
2020-06-22 18:01 ` Roman Gushchin
2020-06-22 18:09 ` Shakeel Butt
2020-06-22 18:25 ` Roman Gushchin
2020-06-22 18:38 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 11/19] mm: memcg/slab: move memcg_kmem_bypass() to memcontrol.h Roman Gushchin
2020-06-20 1:19 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 12/19] mm: memcg/slab: use a single set of kmem_caches for all accounted allocations Roman Gushchin
2020-06-22 16:56 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 13/19] mm: memcg/slab: simplify memcg cache creation Roman Gushchin
2020-06-22 17:29 ` Shakeel Butt
2020-06-22 17:40 ` Roman Gushchin
2020-06-22 18:03 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 14/19] mm: memcg/slab: remove memcg_kmem_get_cache() Roman Gushchin
2020-06-22 18:42 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 15/19] mm: memcg/slab: deprecate slab_root_caches Roman Gushchin
2020-06-22 17:36 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 16/19] mm: memcg/slab: remove redundant check in memcg_accumulate_slabinfo() Roman Gushchin
2020-06-22 17:32 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 17/19] mm: memcg/slab: use a single set of kmem_caches for all allocations Roman Gushchin
2020-06-17 23:35 ` Andrew Morton
2020-06-18 0:35 ` Roman Gushchin
2020-06-18 7:33 ` Vlastimil Babka
2020-06-18 19:54 ` Roman Gushchin
2020-06-22 19:21 ` Shakeel Butt
2020-06-22 20:37 ` Roman Gushchin
2020-06-22 21:04 ` Shakeel Butt
2020-06-22 21:13 ` Roman Gushchin
2020-06-22 21:28 ` Shakeel Butt
2020-06-22 21:58 ` Roman Gushchin
2020-06-22 22:05 ` Shakeel Butt
2020-06-08 23:06 ` [PATCH v6 18/19] kselftests: cgroup: add kernel memory accounting tests Roman Gushchin
2020-06-17 1:46 ` [PATCH v6 00/19] The new cgroup slab memory controller Shakeel Butt
2020-06-17 2:41 ` Roman Gushchin
2020-06-17 3:05 ` Shakeel Butt
2020-06-17 3:32 ` Roman Gushchin
2020-06-17 11:24 ` Vlastimil Babka
2020-06-17 14:31 ` Mel Gorman
2020-06-20 0:57 ` Roman Gushchin
2020-06-18 1:29 ` Roman Gushchin
2020-06-18 8:43 ` Jesper Dangaard Brouer
2020-06-18 9:31 ` Jesper Dangaard Brouer
2020-06-19 1:30 ` Roman Gushchin [this message]
2020-06-19 8:32 ` Jesper Dangaard Brouer
2020-06-19 1:27 ` Roman Gushchin
2020-06-19 9:39 ` Jesper Dangaard Brouer
2020-06-19 18:47 ` Roman Gushchin
2020-06-18 1:18 ` Roman Gushchin
2020-06-18 9:27 ` Mike Rapoport
2020-06-18 20:43 ` Roman Gushchin
2020-06-21 22:57 ` Qian Cai
2020-06-21 23:34 ` Roman Gushchin
2020-06-21 23:53 ` Qian Cai
2020-06-22 3:07 ` Roman Gushchin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200619013013.GE135965@carbon.dhcp.thefacebook.com \
--to=guro@fb.com \
--cc=akpm@linux-foundation.org \
--cc=brouer@redhat.com \
--cc=cl@linux.com \
--cc=hannes@cmpxchg.org \
--cc=kernel-team@fb.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mgorman@techsingularity.net \
--cc=mhocko@kernel.org \
--cc=shakeelb@google.com \
--cc=vbabka@suse.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).