All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=794534

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=5a9f32441aa7a3d48e6cf39a9f2fcbfb3ff10a621462728b940b79460a93e08f


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

chromium-rel-win8-dual

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14b60f52040000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14b60f52040000

cc: Alter duration used in prioritizing impl latency in Scheduler.
By khushalsagar@chromium.org · Thu Dec 07 22:46:18 2017
chromium @ ee432e360c202a47eaf132b33d7c88155f52b6d6

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

khushalsagar: Any progress on this bug? It looks like your CL regressed page load time for one page on Win 8. +kouhei, owner of the loading benchmarks: is this worth pursuing?

Hmmm, looking at the traces before and after the change, it looks like we used to skip a main frame *before* the change landed and we don't anymore. So there's 3 main frames before a long running task from "StreamingCompleteOnBackgroundThread". The change in #3 alters the heuristics for deciding whether we skip frames for latency recovery, but I don't think this is one of the cases where we intended to do this.

Brian, anything I might be missing here?

Kushal's patch likely bought some improvements to first meaningful paint and latency on other pages:
https://chromeperf.appspot.com/group_report?rev=522582

The extra main frame could have improved this case if that main frame were to result in the first meaningful paint. Unfortunately there's a race with something that kicks off the ScriptStreamerThread that looks like it eventually results in the "StreamingCompleteOnBackgroundThread". Detecting one case from the other isn't possible from the compositor thread where Kushal's logic lives.

I wonder if we could increase the priority of ScriptStreamerThread instead?
+skyostil

Otherwise, I think we should close as WontFix.

Closing this. skyostil@, feel free to file a bug if there is something we could improve in main thread scheduling from this case.