Skip to contentSkip to conversation

e04ddfd12aa: +9.4% ebizzy.throughput

mm Fengguang Wu 2 participants 2 messages Apr 4, 2014 – Apr 5, 2014 Memory management

lore
Share & tools
lore
#0 lore raw

Hi Davidlohr,

FYI, there are noticeable ebizzy throughput increases on commit e04ddfd12aa03471eff7daf3bc2435c7cea8e21f ("mm: per-thread vma caching")

test case: lkp-st02/micro/ebizzy/200%-100-10

700141f349f2b05 e04ddfd12aa03471eff7daf3b
--------------- -------------------------

     23726 ~ 0%      +9.4%      25955 ~ 0%  TOTAL ebizzy.throughput
      1525 ~ 0%      +9.3%       1667 ~ 0%  TOTAL ebizzy.throughput.per_thread.max
      1428 ~ 0%      +9.1%       1558 ~ 0%  TOTAL ebizzy.throughput.per_thread.min
     14.42 ~ 0%      +8.1%      15.59 ~ 0%  TOTAL ebizzy.time.user
     65.29 ~ 0%      -1.8%      64.12 ~ 0%  TOTAL ebizzy.time.sys
      4290 ~17%     -28.4%       3073 ~22%  TOTAL slabinfo.buffer_head.active_objs
      4313 ~17%     -28.8%       3073 ~22%  TOTAL slabinfo.buffer_head.num_objs
    165372 ~ 0%      +9.2%     180527 ~ 0%  TOTAL vmstat.system.in
      1443 ~ 0%      +8.1%       1560 ~ 0%  TOTAL time.user_time
      6529 ~ 0%      -1.8%       6413 ~ 0%  TOTAL time.system_time

Legend:
~XX% - stddev percent
[+-]XX% - change percent

ebizzy.throughput

26500 ++O-----------O---O------------------------------------------------+

         O   O O O O     O   O O O     O O                                  |
   26000 ++          O             O O      O     O                         |
         |                                 O  O O                           |
         |                                                                  |
   25500 ++                                                                 |
         |                                                                  |
   25000 ++                                                                 |
         |                                                                  |
   24500 ++                                                                 |
         |                                                                  |
         *.*.*.   .*. .*.*.*.   .*.*.       *.*.*.*.                        |
   24000 ++    *.*   *       *.*     *.*.*.*        *.*.*.*.*.*.            |

| *.*.*.*.*.*.*
23500 ++-----------------------------------------------------------------+

[*] bisect-good sample
[O] bisect-bad sample

Thanks,
Fengguang

#1 lore raw
Quoted 4 lines · Fengguang Wu

Hi Davidlohr,

FYI, there are noticeable ebizzy throughput increases on commit e04ddfd12aa03471eff7daf3bc2435c7cea8e21f ("mm: per-thread vma caching")

Cool, thanks for letting me know.