Mac battor power regressions

[Sunny Sachanandani]

Hi perf-sheriffs,

I'm investigating a mac battor power regression (ChromiumPerf/chromium-rel-mac-retina/battor.power_cases / idle:power_avg) which bisect blamed on a revert I made. However, after relanding the CL in question the regression did not go away completely. The reland went in with hundreds of other CLs including this one which renabled battor.tough_video_cases. The benchmark also had a huge improvement before the revert in this commit range. Was this because of disabling battor benchmarks e.g. this CL in the previous run? Any other ideas about what's up with this benchmark?

Thanks!

- Sunny

[Randy Nephew]

We were having problems with the battor, and it looks like there were 5 days without any runs before the regression. It is likely that in that time, multiple power regressions landed. Because of the this, it might be hard to find which CL is the actual culprit since there could be multiple ones. Looking at the dates of the one right before the regression, at the regression, and after I see these:
Time added: 2016-10-02T03:29:47.000Z
Time added: 2016-10-07T22:35:58.000Z - 5 days
Time added: 2016-10-21T16:12:54.000Z - 14 days

I'm guessing because of the issue we had with the BattOrs, we might not be able to easily move forward with this regression.

Disabling of one battor benchmark should not influence the results of other battor benchmarks.

[Sunny Sachanandani]

I'm also curious about what caused the huge improvement on the benchmark (Point ID: 421875 Time added: 2016-09-30T00:07:12.000Z) prior to the regression. That commit range is small so I've started a bisect.

[Randy Nephew]

If you look at the reference build, it also dropped. For BattOr numbers its important to look at the ref build as well as the normal run. If they both move up and down together, then its likely an environmental issue causing power to change (such as the computer running hot and running the fans harder).