Unexpected memory improvements when enabling SequencedWorkerPool -> TaskScheduler |
||||||||
Issue descriptionhttps://chromeperf.appspot.com/group_report?rev=440433 suspected to be thanks to fewer non-critical tasks running before the test concludes (i.e. prioritization of TaskScheduler works). @fdoray to confirm
,
Jan 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8989396201892222384
,
Jan 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8989396127489887120
,
Jan 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8989396084087799920
,
Jan 26 2017
=== BISECT JOB RESULTS === NO Perf regression found, tests failed to produce values Bisect Details Configuration: win_x64_perf_bisect Benchmark : startup.warm.blank_page Metric : first_non_empty_paint_time/first_non_empty_paint_time To Run This Test src/tools/perf/run_benchmark -v --browser=release_x64 --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests startup.warm.blank_page Debug Info https://chromeperf.appspot.com/buildbucket_job_status/8989396127489887120 Is this bisect wrong? https://chromeperf.appspot.com/bad_bisect?try_job_id=6192226050244608 | 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 Tests>AutoBisect. Thank you!
,
Jan 26 2017
There's a problem with that benchmark, startup.warm.* gives no values with --pageset-repeat=1, crbug.com/677843 I can change the recipe to special case this for now and run 2 iterations if you want.
,
Jan 26 2017
sgtm. I would really like to prove quickly that this performance improvement is due to my CL and I can't repro locally.
,
Jan 26 2017
Started bisect job https://chromeperf.appspot.com/buildbucket_job_status/8989368466650361008
,
Jan 26 2017
=== Auto-CCing suspected CL author gab@chromium.org === Hi gab@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 : gab Commit : 53a4e4af0525c0112bc6b3d03c9f4d2534212830 Date : Thu Dec 22 16:07:49 2016 Subject: Enable BrowserScheduler.RedirectSequencedWorkerPools experiment on trunk. Bisect Details Configuration: win_x64_perf_bisect Benchmark : startup.warm.blank_page Metric : first_non_empty_paint_time/first_non_empty_paint_time Change : 60.56% | 408.222222222 -> 159.444444444 Revision Result N chromium@440399 408.222 +- 714.332 9 good chromium@440425 305.667 +- 469.555 9 good chromium@440432 262.667 +- 233.855 9 good chromium@440433 166.667 +- 19.883 6 bad <-- chromium@440434 204.778 +- 379.299 9 bad chromium@440435 158.667 +- 17.5879 6 bad chromium@440438 165.5 +- 32.9166 14 bad chromium@440450 159.444 +- 21.914 9 bad To Run This Test src/tools/perf/run_benchmark -v --browser=release_x64 --output-format=chartjson --upload-results --pageset-repeat=2 --also-run-disabled-tests startup.warm.blank_page Debug Info https://chromeperf.appspot.com/buildbucket_job_status/8989368466650361008 Is this bisect wrong? https://chromeperf.appspot.com/bad_bisect?try_job_id=4926219679694848 | 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 Tests>AutoBisect. Thank you!
,
Jan 27 2017
=== Auto-CCing suspected CL author perezju@chromium.org === Hi perezju@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 : perezju Commit : 4ee802c50c5c2a316d3209c89c840202750ee9f2 Date : Thu Dec 22 15:21:25 2016 Subject: [system health] Disable browse:news:nytimes on android Bisect Details Configuration: android_nexus6_perf_bisect Benchmark : v8.browsing_mobile_ignition Metric : memory:chrome:renderer_processes:reported_by_chrome:v8:heap:allocated_objects_size_avg/memory:chrome:renderer_processes:reported_by_chrome:v8:heap:allocated_objects_size_avg Change : 13.72% | 11424972.4698 -> 9857920.889 Revision Result N chromium@440379 11424972 +- 574968 6 good chromium@440406 11347859 +- 624360 6 good chromium@440420 11226752 +- 449646 6 good chromium@440422 11333225 +- 215929 6 good chromium@440423 9796001 +- 499050 6 bad <-- chromium@440424 9839986 +- 368416 6 bad chromium@440427 9803156 +- 245075 6 bad chromium@440433 9857921 +- 481918 6 bad To Run This Test src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests v8.browsing_mobile_ignition Debug Info https://chromeperf.appspot.com/buildbucket_job_status/8989396084087799920 Is this bisect wrong? https://chromeperf.appspot.com/bad_bisect?try_job_id=5312615607173120 | 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 Tests>AutoBisect. Thank you!
,
Jan 27 2017
=== BISECT JOB RESULTS === Perf regression found with culprit Suspected Commit Author : perezju Commit : 4ee802c50c5c2a316d3209c89c840202750ee9f2 Date : Thu Dec 22 15:21:25 2016 Subject: [system health] Disable browse:news:nytimes on android Bisect Details Configuration: android_nexus9_perf_bisect Benchmark : v8.browsing_mobile Metric : memory:chrome:renderer_processes:reported_by_chrome:v8:heap:allocated_objects_size_max/memory:chrome:renderer_processes:reported_by_chrome:v8:heap:allocated_objects_size_max Change : 19.63% | 31966420.1212 -> 25690502.2667 Revision Result N chromium@440393 31966420 +- 1903284 6 good chromium@440421 30930137 +- 2671509 9 good chromium@440422 31216264 +- 896032 6 good chromium@440423 26192172 +- 1981429 6 bad <-- chromium@440425 26455167 +- 5775009 9 bad chromium@440428 25316617 +- 1379785 6 bad chromium@440435 25834123 +- 1238577 6 bad chromium@440449 25379741 +- 2069046 6 bad chromium@440504 25690502 +- 2068264 6 bad To Run This Test src/tools/perf/run_benchmark -v --browser=android-chromium --output-format=chartjson --upload-results --pageset-repeat=1 --also-run-disabled-tests v8.browsing_mobile Debug Info https://chromeperf.appspot.com/buildbucket_job_status/8989396201892222384 Is this bisect wrong? https://chromeperf.appspot.com/bad_bisect?try_job_id=5184167161102336 | 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 Tests>AutoBisect. Thank you!
,
Jan 27 2017
You seem to be alerting on an average, rather that on individual pages. This will always trigger false alerts when stories are enabled/disabled. +sullivan
,
Jan 27 2017
,
Jan 27 2017
,
Feb 8 2017
@fdoray: per our offline chat, this was a red herring right? Care to update this bug and the TRIM email thread?
,
Apr 10 2017
Conclusions: - No memory improvement. - Significant startup.warm.blank_page/first_non_empty_paint_time improvement https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg372WtQkM (see comment #9). Could be because TaskScheduler doesn't block https://chromium.googlesource.com/chromium/src/+/6710dbd5124741db004176153c87c609435e7788/base/threading/simple_thread.cc#40 when creating its threads. |
||||||||
►
Sign in to add a comment |
||||||||
Comment 1 by gab@chromium.org
, Jan 18 2017