All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=808650

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=08d25aa740c3fc8abb18c8856060d27a727677eecb0d5bc40e69c2d8142c13be


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

chromium-rel-win10

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/145c90a6840000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/145c90a6840000

Surface synchronization: Reduce synchronization deadline for navigation by fsamuel@chromium.org
chromium @ b46e0efc8062d290e0657c12f22d0453dfd94188

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

Reducing the synchronization deadline causes a regression in timeToFirstPaint while we were expecting the opposite to happen... so weird. Need to investigate.

+cblume, ericrk FYI:

I think we now understand the underlying cause of this regression: allocating a new surface on every navigation added a round trip to from the renderer main thread to the browser UI thread and back to the renderer main thread which incurs queuing delay in both browser UI and renderer main (both busy threads during page loads).

When we had higher synchronization deadlines, some of that cost was offset by doing less work in the browser UI thread (since UI frames were blocked). When the deadline was reduced, we got less jank in browser UI but we slowed down loading slightly.

This problem will hopefully go away in M67 when we hope to introduce child allocated surface IDs. This will avoid the round trip to allocate a surface ID,
and allow the renderer to make forward progress sooner.

Thus, I'm moving this to M67.

 Issue 808649  has been merged into this issue.

 Issue 808616  has been merged into this issue.

 Issue 806797  has been merged into this issue.

Since we will not address this immediately, I'm removing the M67 and Performance Sheriff lable.

This metric is more noisy than before, but it definitely has improved. Marking as FIXED now.