New issue
Advanced search Search tips

Issue 703239 link

Starred by 1 user

Issue metadata

Status: Started
Owner:
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 3
Type: Bug-Regression



Sign in to add a comment

15.8% regression in audio_converter/convert_fifo_only - media_perftests at 457707:457752

Project Member Reported by chcunningham@google.com, Mar 20 2017

Issue description

No ref graph
 
All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=703239

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgjO72sgoM


Bot(s) for this bug's original alert(s):

chromium-rel-win7-gpu-ati
Project Member

Comment 3 by 42576172...@developer.gserviceaccount.com, Mar 20 2017


=== BISECT JOB RESULTS ===
NO Perf regression found

Bisect Details
  Configuration: winx64ati_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_converter/convert_fifo_only

Revision             Result                  N
chromium@457706      1982350 +- 121205       21      good
chromium@457752      1988653 +- 57747.5      21      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984582451696261664

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5887782251134976


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Both values match the pre-regression numbers. Will push out the end revision and re-run.
Project Member

Comment 6 by 42576172...@developer.gserviceaccount.com, Mar 20 2017


=== BISECT JOB RESULTS ===
NO Perf regression found

Bisect Details
  Configuration: winx64ati_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_converter/convert_fifo_only

Revision             Result                  N
chromium@457706      1997970 +- 50831.0      21      good
chromium@457900      1996526 +- 74835.3      21      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8984578087511688624

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5429696071729152


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Status: WontFix (was: Untriaged)
Again, both numbers in post-regression range. No ref graph to compare to, but this looks like noise - closing.
Project Member

Comment 9 by 42576172...@developer.gserviceaccount.com, Apr 11 2017

Cc: chiniforooshan@chromium.org
Owner: chiniforooshan@chromium.org

=== Auto-CCing suspected CL author chiniforooshan@chromium.org ===

Hi chiniforooshan@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : chiniforooshan
  Commit : 6e4c507dc612077518d242eabcfae9622601d7e6
  Date   : Fri Mar 17 07:56:56 2017
  Subject: memory-infra: Finish moving to Mojo (3nd attempt)

Bisect Details
  Configuration: winx64ati_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_converter/convert_fifo_only
  Change       : 5.85% | 1593235.17779 -> 1686481.91911

Revision             Result                  N
chromium@457706      1593235 +- 12334.7      6      good
chromium@457710      1589056 +- 18135.6      6      good
chromium@457711      1583796 +- 12425.1      6      good
chromium@457712      1682613 +- 7685.1       6      bad       <--
chromium@457713      1688031 +- 9278.96      6      bad
chromium@457719      1688676 +- 6809.11      6      bad
chromium@457731      1687445 +- 15106.3      6      bad
chromium@457755      1693223 +- 14219.0      6      bad
chromium@457803      1677754 +- 11803.7      6      bad
chromium@457900      1686482 +- 21070.8      6      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8982615958597668848

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5429696071729152


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Status: Assigned (was: WontFix)
Hi chiniforooshan@, ptal. I'm not sure who kicked off the last bisect, but the jump at your revision is pretty substantial. Is is reasonable to suspect your change?
Status: Started (was: Assigned)
Seeing the bisect result, it is perfectly reasonable to suspect my change :) But it's puzzling how this metric can be related to Mojofication of memory-infra! I will look into this later today.
I haven't found anything yet, but, mysteriously, the graph went back up to where it was before my CL, last week:

https://chromeperf.appspot.com/report?sid=4d41e0480e13e1da6cb7b7cf7d08117f1129c7b84554502da484b30ca2a6c1eb

I'll start a bisect to find out which CL fixed this. That may give us a clue.
Project Member

Comment 16 by 42576172...@developer.gserviceaccount.com, Apr 19 2017

Cc: primiano@chromium.org
Owner: primiano@chromium.org

=== Auto-CCing suspected CL author primiano@chromium.org ===

Hi primiano@chromium.org, the bisect results pointed to your CL, please take a look at the
results.


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : primiano
  Commit : 57cee813cd7d1c400275a5cb8e95612a4fae0fd4
  Date   : Wed Apr 12 11:15:32 2017
  Subject: memory-infra: Switch to MemoryPeakDetector and simplify MemoryDumpScheduler

Bisect Details
  Configuration: winx64ati_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_converter/convert_fifo_only
  Change       : 5.56% | 1689510.05262 -> 1595632.626

Revision             Result                  N
chromium@463938      1689510 +- 8384.43      6      good
chromium@463969      1684943 +- 19935.5      6      good
chromium@463985      1696077 +- 20362.2      6      good
chromium@463989      1696123 +- 14905.9      6      good
chromium@463990      1684614 +- 12716.6      6      good
chromium@463991      1593454 +- 7720.93      6      bad       <--
chromium@463993      1594238 +- 18827.6      6      bad
chromium@464000      1595633 +- 13667.1      6      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8981873735744757952

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=6463512731189248


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Project Member

Comment 17 by 42576172...@developer.gserviceaccount.com, Apr 19 2017


=== BISECT JOB RESULTS ===
Perf regression found with culprit

Suspected Commit
  Author : primiano
  Commit : 57cee813cd7d1c400275a5cb8e95612a4fae0fd4
  Date   : Wed Apr 12 11:15:32 2017
  Subject: memory-infra: Switch to MemoryPeakDetector and simplify MemoryDumpScheduler

Bisect Details
  Configuration: winx64ati_perf_bisect
  Benchmark    : media_perftests
  Metric       : audio_converter/convert_fifo_only
  Change       : 5.83% | 1682586.43364 -> 1584504.56099

Revision             Result                  N
chromium@463938      1682586 +- 8816.66      6      good
chromium@463969      1684711 +- 20973.8      6      good
chromium@463985      1685423 +- 14166.5      6      good
chromium@463989      1692041 +- 10074.5      6      good
chromium@463990      1679833 +- 9887.03      6      good
chromium@463991      1591983 +- 14195.0      6      bad       <--
chromium@463993      1587324 +- 20373.6      6      bad
chromium@464000      1584505 +- 13701.1      6      bad

To Run This Test
  .\src\out\Release_x64\media_perftests.exe --single-process-tests

Debug Info
  https://chromeperf.appspot.com/buildbucket_job_status/8981872972943772352

Is this bisect wrong?
  https://chromeperf.appspot.com/bad_bisect?try_job_id=5133734552010752


| O O | Visit http://www.chromium.org/developers/speed-infra/perf-bug-faq
|  X  | for more information addressing perf regression bugs. For feedback,
| / \ | file a bug with component Speed>Bisection.  Thank you!
Labels: -Pri-2 Pri-3
Owner: chiniforooshan@chromium.org
OK, so it was something my CL did that primiano's CL undid it. I'll continue investigating this on a lower priority since it looks like the bug is "fixed".
Uh? 
Honestly this very suspiciously smells like  Issue 656729  (the startup trace config being leaked). I can't tell how either mine or chiniforooshan@ change could affect a media_perftest.
That bimodal pattern in the chart in #0 is very suspicious and smells to me as if tracing is flakily on. Which is odd, because the code of media_perftest doesn't seem to make use any use of tracing.
Project Member

Comment 20 by sheriffbot@chromium.org, Jul 21 2017

Labels: Hotlist-Google

Sign in to add a comment