Skip to content

Reduce amount of time kswapd sleeps prematurely

Mel Gorman linux-mm lkml 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest3 patches 20 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/3mm, vmscan: fix zone balance check in prepare_kswapd_sleepmm/vmscan.c+3 -36 replies
2/3mm, vmscan: Only clear pgdat congested/dirty/writeback state when balancedmm/vmscan.c+11 -90 replies
3/3mm, vmscan: Prevent kswapd sleeping prematurely due to mismatched classzone_idxmm/memory_hotplug.c mm/vmscan.c+66 -549 replies

Reviews trailers per patch

2 of 3 patches have a Reviewed-by or Acked-by.

PatchHillf DantonVlastimil Babka
1/3A-b
2/3
3/3A-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v1 thread

Cover letter [PATCH 0/3] Reduce amount of time kswapd sleeps prematurely

This patchset is based on mmots as of Feb 9th, 2016. The baseline is important as there are a number of kswapd-related fixes in that tree and a comparison against v4.10-rc7 would be almost meaningless as a result.

The series is unusual in that the first patch fixes one problem and introduces a host of other issues and is incomplete. It was not developed by me but it appears to have gotten lost so I picked it up and added to the changelog. Patch 2 makes a minor modification that is worth considering on its own but leaves the kernel in a state where it behaves badly. It's not until patch 3 that there is an improvement against baseline.

This was mostly motivated by examining Chris Mason's "simoop" benchmark which puts the VM under similar pressure to HADOOP. It has been reported that the benchmark has regressed severely during the last number of releases. While I cannot reproduce all the same problems Chris experienced due to hardware limitations, there was a number of problems on a 2-socket machine with a single disk.

                                         4.10.0-rc7            4.10.0-rc7
                                     mmots-20170209       keepawake-v1r25
Amean    p50-Read             22325202.49 (  0.00%) 22092755.48 (  1.04%)
Amean    p95-Read             26102988.80 (  0.00%) 26101849.04 (  0.00%)
Amean    p99-Read             30935176.53 (  0.00%) 29746220.52 (  3.84%)
Amean    p50-Write                 976.44 (  0.00%)      952.73 (  2.43%)
Amean    p95-Write               15471.29 (  0.00%)     3140.27 ( 79.70%)
Amean    p99-Write               35108.62 (  0.00%)     8843.73 ( 74.81%)
Amean    p50-Allocation          76382.61 (  0.00%)    76349.22 (  0.04%)
Amean    p95-Allocation         127777.39 (  0.00%)   108630.26 ( 14.98%)
Amean    p99-Allocation         187937.39 (  0.00%)   139094.26 ( 25.99%)

These are latencies. Read/write are threads reading fixed-size random blocks from a simulated database. The allocation latency is mmaping and faulting regions of memory. The p50, 95 and p99 reports the worst latencies for 50% of the samples, 95% and 99% respectively.

For example, the report indicates that while the test was running 99% of writes completed 74.81% faster. It's worth noting that on a UMA machine that no difference in performance with simoop was observed so milage will vary.

On UMA, there was a notable difference in the "stutter" benchmark which measures the latency of mmap while large files are being copied. This has been used as a proxy measure for desktop jitter while large amounts of IO were taking place

                            4.10.0-rc7            4.10.0-rc7
                        mmots-20170209          keepawake-v1
Min         mmap      6.3847 (  0.00%)      5.9785 (  6.36%)
1st-qrtle   mmap      7.6310 (  0.00%)      7.4086 (  2.91%)
2nd-qrtle   mmap      9.9959 (  0.00%)      7.7052 ( 22.92%)
3rd-qrtle   mmap     14.8180 (  0.00%)      8.5895 ( 42.03%)
Max-90%     mmap     15.8397 (  0.00%)     13.6974 ( 13.52%)
Max-93%     mmap     16.4268 (  0.00%)     14.3175 ( 12.84%)
Max-95%     mmap     18.3295 (  0.00%)     16.9233 (  7.67%)
Max-99%     mmap     24.2042 (  0.00%)     20.6182 ( 14.82%)
Max         mmap    255.0688 (  0.00%)    265.5818 ( -4.12%)
Mean        mmap     11.2192 (  0.00%)      9.1811 ( 18.17%)

Latency is measured in milliseconds and indicates that 99% of mmap operations complete 14.82% faster and are 18.17% faster on average with these patches applied.

 mm/memory_hotplug.c |   2 +-
 mm/vmscan.c         | 128 +++++++++++++++++++++++++++++-----------------------
 2 files changed, 72 insertions(+), 58 deletions(-)
Signature · 23 lines
2.11.0

Mel Gorman (2):
  mm, vmscan: Only clear pgdat congested/dirty/writeback state when
    balanced
  mm, vmscan: Prevent kswapd sleeping prematurely due to mismatched
    classzone_idx

Shantanu Goel (1):
  mm, vmscan: fix zone balance check in prepare_kswapd_sleep

 mm/memory_hotplug.c |   2 +-
 mm/vmscan.c         | 128 +++++++++++++++++++++++++++++-----------------------
 2 files changed, 72 insertions(+), 58 deletions(-)

-- 
2.11.0

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>