All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=853498

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=487b4d2165d4604d3c67b8f49077e4bc40283295fe844319e124661dbebc0e11


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

chromium-rel-win10

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/12fa8825240000

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/12fa8825240000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/12fde2dd240000

^^^ New run with wider range. 

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/12fde2dd240000

Over to speed > bisection. Graph shows clear sustained regression vs the ref, but bisect is not finding it. 

Very similar to (possibly same root cause)  Issue 853499 

+dtu

Dave, can you help them out?

Based on the perf dashboard graph, we're expecting a jump from around 330 to 380.
I looked at the means of the two Pinpoint jobs, and got:

comment 3 (medians): 579 477 436
comment 3 (means): 979 756 740
comment 6 (medians): 1291 762
comment 6 (means): 1447 1173

These aren't at all close to what we saw on the dashboard. And in fact they show an improvement, rather than a regression. Simon's first thought was that maybe the stories aren't independent, i.e. some data is leaking between the webpages in the benchmark. By default Pinpoint bisects only run the one page. So I'll try running the same job with the full benchmark with all pages.

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/11b9045f240000

I also thought that maybe the hardware is different, but I checked and found that the perf waterfall and Pinpoint jobs ran with similar CPUs and GPUs.

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/11b9045f240000

The results with the full benchmark (comment 12) are much more consistent with the "before" numbers we saw on perf dashboard. I'll try expanding the end of the range and run again.

I also notice that the ref build also jumped at the "after" point, and the graph in  issue 853499  shows the regression on the following point, which are a few extra reasons to expand the end of the range.

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14ad7af7240000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14ad7af7240000

Field trial for AudioService experiment by olka@chromium.org
https://chromium.googlesource.com/chromium/src/+/8d891be67dac6be0f01a3c91e56ca9729101782c
341.3 → 416.8 (+75.48)

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions