From: "Huang\, Ying" <ying.huang@linux.intel.com>
To: Hugh Dickins <hughd@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Josef Bacik <jbacik@fb.com>, Yu Zhao <yuzhao@google.com>,
linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [PATCH] tmpfs: avoid a little creat and stat slowdown
Date: Fri, 13 Nov 2015 16:33:04 +0800 [thread overview]
Message-ID: <8737wafge7.fsf@yhuang-dev.intel.com> (raw)
In-Reply-To: <alpine.LSU.2.11.1511081543590.14116@eggly.anvils> (Hugh Dickins's message of "Sun, 8 Nov 2015 16:15:51 -0800 (PST)")
Hugh Dickins <hughd@google.com> writes:
> On Wed, 4 Nov 2015, Huang, Ying wrote:
>> Hugh Dickins <hughd@google.com> writes:
>>
>> > LKP reports that v4.2 commit afa2db2fb6f1 ("tmpfs: truncate prealloc
>> > blocks past i_size") causes a 14.5% slowdown in the AIM9 creat-clo
>> > benchmark.
>> >
>> > creat-clo does just what you'd expect from the name, and creat's O_TRUNC
>> > on 0-length file does indeed get into more overhead now shmem_setattr()
>> > tests "0 <= 0" instead of "0 < 0".
>> >
>> > I'm not sure how much we care, but I think it would not be too VW-like
>> > to add in a check for whether any pages (or swap) are allocated: if none
>> > are allocated, there's none to remove from the radix_tree. At first I
>> > thought that check would be good enough for the unmaps too, but no: we
>> > should not skip the unlikely case of unmapping pages beyond the new EOF,
>> > which were COWed from holes which have now been reclaimed, leaving none.
>> >
>> > This gives me an 8.5% speedup: on Haswell instead of LKP's Westmere,
>> > and running a debug config before and after: I hope those account for
>> > the lesser speedup.
>> >
>> > And probably someone has a benchmark where a thousand threads keep on
>> > stat'ing the same file repeatedly: forestall that report by adjusting
>> > v4.3 commit 44a30220bc0a ("shmem: recalculate file inode when fstat")
>> > not to take the spinlock in shmem_getattr() when there's no work to do.
>> >
>> > Reported-by: Ying Huang <ying.huang@linux.intel.com>
>> > Signed-off-by: Hugh Dickins <hughd@google.com>
>>
>> Hi, Hugh,
>>
>> Thanks a lot for your support! The test on LKP shows that this patch
>> restores a big part of the regression! In following list,
>>
>> c435a390574d012f8d30074135d8fcc6f480b484: is parent commit
>> afa2db2fb6f15f860069de94a1257db57589fe95: is the first bad commit has
>> performance regression.
>> 43819159da2b77fedcf7562134d6003dccd6a068: is the fixing patch
>
> Hi Ying,
>
> Thank you, for reporting, and for trying out the patch (which is now
> in Linus's tree as commit d0424c429f8e0555a337d71e0a13f2289c636ec9).
>
> But I'm disappointed by the result: do I understand correctly,
> that afa2db2fb6f1 made a -12.5% change, but the fix still -5.6%
> from your parent comparison point?
Yes.
> If we value that microbenchmark
> at all (debatable), I'd say that's not good enough.
I think that is a good improvement.
> It does match with my own rough measurement, but I'd been hoping
> for better when done in a more controlled environment; and I cannot
> explain why "truncate prealloc blocks past i_size" creat-clo performance
> would not be fully corrected by "avoid a little creat and stat slowdown"
> (unless either patch adds subtle icache or dcache displacements).
>
> I'm not certain of how you performed the comparison. Was the
> c435a390574d tree measured, then patch afa2db2fb6f1 applied on top
> of that and measured, then patch 43819159da2b applied on top of that
> and measured? Or were there other intervening changes, which could
> easily add their own interference?
c435a390574d is the direct parent of afa2db2fb6f1 in its original git.
43819159da2b is your patch applied on top of v4.3-rc7. The comparison
of 43819159da2b with v4.3-rc7 is as follow:
=========================================================================================
compiler/cpufreq_governor/kconfig/rootfs/tbox_group/test/testcase/testtime:
gcc-4.9/performance/x86_64-rhel/debian-x86_64-2015-02-07.cgz/lkp-wsx02/creat-clo/aim9/300s
commit:
32b88194f71d6ae7768a29f87fbba454728273ee
43819159da2b77fedcf7562134d6003dccd6a068
32b88194f71d6ae7 43819159da2b77fedcf7562134
---------------- --------------------------
%stddev %change %stddev
\ | \
475224 A+- 1% +11.9% 531968 A+- 1% aim9.creat-clo.ops_per_sec
10469094 A+-201% -52.3% 4998529 A+-130% latency_stats.avg.nfs_wait_on_request.nfs_updatepage.nfs_write_end.generic_perform_write.__generic_file_write_iter.generic_file_write_iter.nfs_file_write.__vfs_write.vfs_write.SyS_write.entry_SYSCALL_64_fastpath
18852332 A+-223% -73.5% 4998529 A+-130% latency_stats.max.nfs_wait_on_request.nfs_updatepage.nfs_write_end.generic_perform_write.__generic_file_write_iter.generic_file_write_iter.nfs_file_write.__vfs_write.vfs_write.SyS_write.entry_SYSCALL_64_fastpath
21758590 A+-199% -77.0% 4998529 A+-130% latency_stats.sum.nfs_wait_on_request.nfs_updatepage.nfs_write_end.generic_perform_write.__generic_file_write_iter.generic_file_write_iter.nfs_file_write.__vfs_write.vfs_write.SyS_write.entry_SYSCALL_64_fastpath
4817724 A+- 0% +9.6% 5280303 A+- 1% proc-vmstat.numa_hit
4812582 A+- 0% +9.7% 5280287 A+- 1% proc-vmstat.numa_local
8499767 A+- 4% +14.2% 9707953 A+- 4% proc-vmstat.pgalloc_normal
8984075 A+- 0% +10.4% 9919044 A+- 1% proc-vmstat.pgfree
9.22 A+- 8% +27.4% 11.75 A+- 9% sched_debug.cfs_rq[0]:/.nr_spread_over
2667 A+- 63% +90.0% 5068 A+- 37% sched_debug.cfs_rq[20]:/.min_vruntime
152513 A+-272% -98.5% 2306 A+- 48% sched_debug.cfs_rq[21]:/.min_vruntime
477.36 A+- 60% +128.6% 1091 A+- 60% sched_debug.cfs_rq[27]:/.exec_clock
4.00 A+-112% +418.8% 20.75 A+- 67% sched_debug.cfs_rq[28]:/.util_avg
1212 A+- 80% +195.0% 3577 A+- 48% sched_debug.cfs_rq[29]:/.exec_clock
8119 A+- 53% -60.4% 3217 A+- 26% sched_debug.cfs_rq[2]:/.min_vruntime
584.80 A+- 65% -60.0% 234.06 A+- 13% sched_debug.cfs_rq[30]:/.exec_clock
4245 A+- 27% -42.8% 2429 A+- 24% sched_debug.cfs_rq[30]:/.min_vruntime
0.00 A+- 0% +Inf% 2.25 A+- 72% sched_debug.cfs_rq[44]:/.util_avg
1967 A+- 39% +72.0% 3384 A+- 15% sched_debug.cfs_rq[61]:/.min_vruntime
1863 A+- 43% +99.2% 3710 A+- 33% sched_debug.cfs_rq[72]:/.min_vruntime
0.78 A+-336% -678.6% -4.50 A+--33% sched_debug.cpu#12.nr_uninterruptible
10686 A+- 49% +77.8% 19002 A+- 34% sched_debug.cpu#15.nr_switches
5256 A+- 50% +79.0% 9410 A+- 34% sched_debug.cpu#15.sched_goidle
-2.00 A+--139% -225.0% 2.50 A+- 44% sched_debug.cpu#21.nr_uninterruptible
-1.78 A+--105% -156.2% 1.00 A+-141% sched_debug.cpu#23.nr_uninterruptible
45017 A+-132% -76.1% 10741 A+- 30% sched_debug.cpu#24.nr_load_updates
2216 A+- 14% +73.3% 3839 A+- 63% sched_debug.cpu#35.nr_switches
2223 A+- 14% +73.0% 3845 A+- 63% sched_debug.cpu#35.sched_count
1030 A+- 13% +79.1% 1845 A+- 66% sched_debug.cpu#35.sched_goidle
2.00 A+- 40% +37.5% 2.75 A+- 82% sched_debug.cpu#46.nr_uninterruptible
907.11 A+- 67% +403.7% 4569 A+- 75% sched_debug.cpu#59.ttwu_count
-4.56 A+--41% -94.5% -0.25 A+--714% sched_debug.cpu#64.nr_uninterruptible
So you patch improved 11.9% from its base v4.3-rc7. I think other
difference are caused by other changes. Sorry for confusing.
Best Regards,
Huang, Ying
> Hugh
>
>>
>> =========================================================================================
>> compiler/cpufreq_governor/kconfig/rootfs/tbox_group/test/testcase/testtime:
>> gcc-4.9/performance/x86_64-rhel/debian-x86_64-2015-02-07.cgz/lkp-wsx02/creat-clo/aim9/300s
>>
>> commit:
>> c435a390574d012f8d30074135d8fcc6f480b484
>> afa2db2fb6f15f860069de94a1257db57589fe95
>> 43819159da2b77fedcf7562134d6003dccd6a068
>>
>> c435a390574d012f afa2db2fb6f15f860069de94a1 43819159da2b77fedcf7562134
>> ---------------- -------------------------- --------------------------
>> %stddev %change %stddev %change %stddev
>> \ | \ | \
>> 563556 A+- 1% -12.5% 493033 A+- 5% -5.6% 531968 A+- 1% aim9.creat-clo.ops_per_sec
>> 11836 A+- 7% +11.4% 13184 A+- 7% +15.0% 13608 A+- 5% numa-meminfo.node1.SReclaimable
>> 10121526 A+- 3% -12.1% 8897097 A+- 5% -4.1% 9707953 A+- 4% proc-vmstat.pgalloc_normal
>> 9.34 A+- 4% -11.4% 8.28 A+- 3% -4.8% 8.88 A+- 2% time.user_time
>> 3480 A+- 3% -2.5% 3395 A+- 1% -28.5% 2488 A+- 3% vmstat.system.cs
>> 203275 A+- 17% -6.8% 189453 A+- 5% -34.4% 133352 A+- 11% cpuidle.C1-NHM.usage
>> 8081280 A+-129% -93.3% 538377 A+- 97% +31.5% 10625496 A+-106% cpuidle.C1E-NHM.time
>> 3144 A+- 58% +619.0% 22606 A+- 56% +903.9% 31563 A+- 0% numa-vmstat.node0.numa_other
>> 2958 A+- 7% +11.4% 3295 A+- 7% +15.0% 3401 A+- 5% numa-vmstat.node1.nr_slab_reclaimable
>> 45074 A+- 5% -43.4% 25494 A+- 57% -68.7% 14105 A+- 2% numa-vmstat.node2.numa_other
>> 56140 A+- 0% +0.0% 56158 A+- 0% -94.4% 3120 A+- 0% slabinfo.Acpi-ParseExt.active_objs
>> 1002 A+- 0% +0.0% 1002 A+- 0% -92.0% 80.00 A+- 0% slabinfo.Acpi-ParseExt.active_slabs
>> 56140 A+- 0% +0.0% 56158 A+- 0% -94.4% 3120 A+- 0% slabinfo.Acpi-ParseExt.num_objs
>> 1002 A+- 0% +0.0% 1002 A+- 0% -92.0% 80.00 A+- 0% slabinfo.Acpi-ParseExt.num_slabs
>> 1079 A+- 5% -10.8% 962.00 A+- 10% -100.0% 0.00 A+- -1% slabinfo.blkdev_ioc.active_objs
>> 1079 A+- 5% -10.8% 962.00 A+- 10% -100.0% 0.00 A+- -1% slabinfo.blkdev_ioc.num_objs
>> 110.67 A+- 39% +74.4% 193.00 A+- 46% +317.5% 462.00 A+- 8% slabinfo.blkdev_queue.active_objs
>> 189.33 A+- 23% +43.7% 272.00 A+- 33% +151.4% 476.00 A+- 10% slabinfo.blkdev_queue.num_objs
>> 1129 A+- 10% -1.9% 1107 A+- 7% +20.8% 1364 A+- 6% slabinfo.blkdev_requests.active_objs
>> 1129 A+- 10% -1.9% 1107 A+- 7% +20.8% 1364 A+- 6% slabinfo.blkdev_requests.num_objs
>> 1058 A+- 3% -10.3% 949.00 A+- 9% -100.0% 0.00 A+- -1% slabinfo.file_lock_ctx.active_objs
>> 1058 A+- 3% -10.3% 949.00 A+- 9% -100.0% 0.00 A+- -1% slabinfo.file_lock_ctx.num_objs
>> 4060 A+- 1% -2.1% 3973 A+- 1% -10.5% 3632 A+- 1% slabinfo.files_cache.active_objs
>> 4060 A+- 1% -2.1% 3973 A+- 1% -10.5% 3632 A+- 1% slabinfo.files_cache.num_objs
>> 10001 A+- 0% -0.3% 9973 A+- 0% -61.1% 3888 A+- 0% slabinfo.ftrace_event_field.active_objs
>> 10001 A+- 0% -0.3% 9973 A+- 0% -61.1% 3888 A+- 0% slabinfo.ftrace_event_field.num_objs
>> 1832 A+- 0% +0.4% 1840 A+- 0% -100.0% 0.00 A+- -1% slabinfo.ftrace_event_file.active_objs
>> 1832 A+- 0% +0.4% 1840 A+- 0% -100.0% 0.00 A+- -1% slabinfo.ftrace_event_file.num_objs
>> 1491 A+- 5% -2.3% 1456 A+- 6% +12.0% 1669 A+- 4% slabinfo.mnt_cache.active_objs
>> 1491 A+- 5% -2.3% 1456 A+- 6% +12.0% 1669 A+- 4% slabinfo.mnt_cache.num_objs
>> 126.33 A+- 19% +10.2% 139.17 A+- 9% -100.0% 0.00 A+- -1% slabinfo.nfs_commit_data.active_objs
>> 126.33 A+- 19% +10.2% 139.17 A+- 9% -100.0% 0.00 A+- -1% slabinfo.nfs_commit_data.num_objs
>> 97.17 A+- 20% -9.1% 88.33 A+- 28% -100.0% 0.00 A+- -1% slabinfo.user_namespace.active_objs
>> 97.17 A+- 20% -9.1% 88.33 A+- 28% -100.0% 0.00 A+- -1% slabinfo.user_namespace.num_objs
>>
>> Best Regards,
>> Huang, Ying
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2015-11-13 8:33 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-10-29 19:13 [PATCH] tmpfs: avoid a little creat and stat slowdown Hugh Dickins
2015-11-04 7:47 ` Huang, Ying
2015-11-09 0:15 ` Hugh Dickins
2015-11-13 8:33 ` Huang, Ying [this message]
2015-11-14 0:51 ` Hugh Dickins
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=8737wafge7.fsf@yhuang-dev.intel.com \
--to=ying.huang@linux.intel.com \
--cc=akpm@linux-foundation.org \
--cc=hughd@google.com \
--cc=jbacik@fb.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=yuzhao@google.com \
/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