All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=598063

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgsLuMpwoM


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

android-galaxy-s5

All of these alerts are over 22,000 revisions. Do we think there's any chance of narrowing this down, or that they're even all the same regression?

+picksi

Looks like there was an outage on the perfbots and we have a memory regression that is too big to bisect. Simon, are we seeing this regression on our other bots?

This benchmark was disabled for a long while ( issue 581147 ) and that's why we have that very long gap of CLs.

There is probably not much to do about this, and when looking at the whole of vm_private_dirty_final_browser (not just a single page) the regression seems to have somewhat recovered:
https://chromeperf.appspot.com/report?sid=6eeb4b64bd09b6286d07c4bce6c4bf51bda38ace4535dfbd9bf4426f2551a01b&start_rev=370073&end_rev=384858&rev=383083

We've been also tracking memory continuously with the memory_healt_plan benchmark, catching and fixing regressions. So I suggest we focus on that for the moment.

+1 to perezju@ in #4
It looks pretty similar in what we had on other benchmarks over time (see for instance https://chromeperf.appspot.com/report?sid=4304b3cc3247ee149d9eee9ac4ba216dc2f9f1d6fe99ff350816c284b3b91bd0&start_rev=371171&end_rev=384522)
Also, if anybody wants to bisect 20k CLs let me know and I will bring popcorn :)