All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=667716

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgm8HVpQoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgm8HssAoM


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

webrtc-win-large-tests

brandtr@: Could these regressions also be related to the shortened duration of the tests?

The shortening of the tests happened at #15118, which gives a pretty clear uptick in the graphs (but not a perf alert!). But it's weird that there is later a downtick, where the encode time goes down. In the downtick blamelist, I found this CL: https://codereview.webrtc.org/2505183003, which I don't fully understand.

kjellander@: Did that CL move the Win bots to dedicated machines, thus explaining the reduced encode time in these tests?

That CL is related to the move (which happened within the same hour), but the CL itself only covers removing DCHECKs, which was expected trigger a large number of improvement alerts (as communicated to the sheriffs mailing list).

Right, but from the CL description it seemed that only Android bots had their DCHECKs removed. But I see in the email that you sent out that the desktop tests were affected too. Thanks for looking into it!