New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 745508 link

Starred by 1 user

Issue metadata

Status: WontFix
Owner:
Closed: Aug 2017
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug-Regression



Sign in to add a comment

38.2% improvement in webrtc_perf_tests at 18976:18976

Project Member Reported by ilnik@chromium.org, Jul 18 2017

Issue description

Chromium roll caused all kind of improvements and regressions around. 6 days later another chromium roll reverted all the changes.

Original roll with improvements:
https://chromium.googlesource.com/external/webrtc/+/f1f9889c447d33dac25bfcfea0cc99f8cc817af0

Revert happened at:
https://chromium.googlesource.com/external/webrtc/+/cfd3c327a8660a990a7fdd9d38f7366393cf40f0

Anyone have any idea, what happened?
 
Project Member

Comment 1 by 42576172...@developer.gserviceaccount.com, Jul 18 2017

All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=745508

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=5e5956dbb6a48ac69f0968d547a41e08f94332b965bf41382a3502d4408872ca


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

webrtc-android-tests-nexus4-lollipop

Comment 2 by ilnik@chromium.org, Jul 18 2017

It's especially sad, because it was like magic: increased psnr, reduced encode and decode time, reduced cpu usage.

Comment 3 by tommi@chromium.org, Jul 18 2017

Can we bisect?

Comment 4 by ilnik@chromium.org, Jul 18 2017

Re 3#: Do you mean rolling parts of chromium roll? Do you know how to do it in the easiest way? 

Comment 5 by tommi@chromium.org, Jul 18 2017

ah, hmm... not sure how to do that. Perhaps kjellander@ could help?
Status: WontFix (was: Assigned)
We don't have bisection support for the perf tests in client.webrtc.perf waterfall (i.e. standalone WebRTC perf tests). Only Chromium Perf waterfalls has bisection support (and can do it into rolls as well).

In this case, I don't think we should spend time on this if it recovered. As we were seeing both regressions and improvements I would guess there were something wrong that traded quality for bandwidth or similar. I suggest we close this.

Sign in to add a comment