Issue metadata
Sign in to add a comment
|
tough_webgl_cases mean_frame_times got very noisy around 9/3/2016 |
||||||||||||||||||||||
Issue descriptionvmiura@ pointed out that the mean_frame_times of the tough_webgl_cases performance tests got very noisy around September 3, 2016: https://chromeperf.appspot.com/report?sid=220fc7bd7a1074610c4caeb6960600ef162ee330cdff793edbbb3b55b0dd99c6&start_rev=415037&end_rev=426150 It's not clear what may have caused this. This regression list seems to be conservative: http://test-results.appspot.com/revision_range?start=415884&end=416454 and it contains no changes that are obviously relevant. vmiura@ suggested perhaps throttling changes by ericrk@ and ccameron@, but this blamelist doesn't contain them. There are a few V8 rolls, which I haven't dug into, but suspect they would not be the cause. The same issue doesn't seem to show up on other platforms, i.e., Windows: https://chromeperf.appspot.com/report?sid=bdde1fe421bcd5434ec9e48f4869d12fe58e45e3970e17923c4cb173a61681ad Does anyone have an idea about what might have changed? The affected tests are pretty simple and don't do much CPU or GPU work.
,
Oct 19 2016
2 V8 rolls: - https://chromium.googlesource.com/v8/v8/+log/31399854..3fd3ba53 - https://chromium.googlesource.com/v8/v8/+log/0e51f51f..31399854 Only interesting (don't know if the benchmark makes use of DataView_s) CL: - https://chromium.googlesource.com/v8/v8/+/97f6266f564e0b8a6f8ecd2f12cfadfccccff511: [turbofan] Migrate remaining DataView builtins to C++.
,
Oct 19 2016
None of the benchmarks should be using DataView. Thanks for taking a look mlippautz@.
,
Oct 19 2016
Sorry, I missed a couple of v8 rolls due to the paged rev list. There is one more interesting commit: https://chromium.googlesource.com/v8/v8/+/ce66eec800e7329b0967b0f8eec7db05b2b22218: [heap] Introduce hard limit for external memory We the incremental marking phase of V8 earlier (64M), but move hard limit for doing a full GC to half of the V8 heap size (700M on 64bit). Previously we would start GCing incrementally at ~200M and not make any progress (which actually is buggy behavior). Can we do a bisect?
,
Oct 19 2016
Kicked off a bisect between revisions 415828 and 416464.
,
Oct 20 2016
,
Oct 21 2016
,
May 1 2017
,
May 2 2018
This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue. Sorry for the inconvenience if the bug really should have been left as Available. For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot
,
May 2 2018
These seem to have settled down by themselves. There have been tons of changes since this bug was reported. https://chromeperf.appspot.com/report?sid=a84d11874fff64121609f33cc88622062606f12df602364f1250b3432c9c9143 |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by kbr@chromium.org
, Oct 19 2016