All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=877247

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=05a0a7b0d23be4b31c0c14be2b995402d0f63222a4d2e4e1b771d6a32550b2ad


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

Android Nexus5 Perf
android-nexus5x-perf

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/1403705e640000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/1403705e640000

Remove Modern flag and hardcode #isChromeModernDesignEnabled to true by twellington@chromium.org
https://chromium.googlesource.com/chromium/src/+/5955e6d67f3799a0cdd9dba7df8fa5460499f99f
5.381e+06 → 5.471e+06 (+8.926e+04)

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

+other folks working on Modern - It looks like there was a regression in GPU memory usage after enabling Modern for the test bots.

+media-mobile owners -- Chrome Modern was a visual redesign of some of Clank's core UI. Does it make sense that changing the toolbar style would cause a regression in GPU memory for this perf test? 

 Issue 877244  has been merged into this issue.

Yeah, it makes sense that a new design might cause a regression. I would hope that this regression is noted in the launch documentation for Modern. We don't want to gradually bloat Chrome over time with a bunch of small regressions. But it is true that this regression is pretty small.

It's not noted anywhere yet (besides this bug). We didn't anticipate any performance regressions and we didn't see any regressions in UMA data during experimentation. Is there a standard way to document this e.g. in the launch bug or in the design doc?

I don't know of one. +Tim maybe does.

Issue 877184 has been merged into this issue.

I captured this perf bot regression in the metrics analysis doc for Modern (draft/work-in-progress): https://uma.googleplex.com/p/chrome/variations/?sid=4d1804556c32d47e8fd55935c8f3eb8a

It's worth noting that we did not see a similar GPU memory regression in the experiment group for the modern variant that we are launching ("ImageLast"). The median GPU memory usage for our experiment group (ramped up in M68) is roughly the same as the control group (Control: 27.5598MB, ImageLast: 27.5582MB) as is the mean memory (screenshot attached).

This alert is for the "image first" Zine card variant* which has the same mean as the other two groups. The median (27.5874MB) is very slightly higher (~0.1% increase from control) but an order of magnitude lower than the 1.2-1.7% found on the perf bot.

UMA link https://uma.googleplex.com/p/chrome/variations/?sid=4d1804556c32d47e8fd55935c8f3eb8a

*The image last variant was enabled on ToT in this CL on Aug 30:  https://chromium-review.googlesource.com/c/chromium/src/+/1195435

Deleted: Screen Shot 2018-09-02 at 9.14.39 AM.png 25.6 KB

	Screen Shot 2018-09-02 at 9.14.39 AM.png 25.6 KB View Download

 Issue 880411  has been merged into this issue.

 Issue 877603  has been merged into this issue.