Filip Andrei Reporter
	Description • 7 years ago

1.Login into Gmail
2.Start Gecko profile
3.Access inbox
4.Share profile


Gecko profile:
https://perfht.ml/2r2bqpe

	Bobby Chien
	Comment 1 • 7 years ago

Neerja, could you help to have first profiling? Thanks.

	Neerja Pancholi[:neerja]
	Comment 2 • 7 years ago

(In reply to Bobby Chien [:bchien] from comment #1)
> Neerja, could you help to have first profiling? Thanks.

Thanks Bobby, I'll look into it.

	Mike Conley (:mconley) (:⚙️)
	Comment 4 • 7 years ago

mstange and I looked at this profile today in The Joy of Profiling Episode 8[1].

Most of the profile is JS bound. Folks from the JS team (perhaps sfink or jandem?) should look at this to see if there's anything new in here that's actionable.

Note that the profile is 2 months old. A lot of work has already gone into optimizing JS for Speedometer in the meantime. It might be worth re-gathering this profile to see the current state of things.

Other observations:

1) Some longer paints seem to be blocked on pushing layers, which flushes draw calls to D2D. OMTP could help here.
2) Bug 1374797 might have helped with this, which is more reason to gather a newer, up-to-date profile.

Can a new profile be gathered, please?

[1]: https://air.mozilla.org/the-joy-of-profiling-episode-8/

	Filip Andrei Reporter
	Comment 5 • 7 years ago

Here is the profile that you requested https://perfht.ml/2hfXjwj and if i should change interval or the buffer let me know.

	Mike Conley (:mconley) (:⚙️)
	Comment 6 • 7 years ago

This profile shows really nice improvements, at least in terms of jank. We're janking a lot less now, which is good!

The remaining jank looks to be JS bound here: https://perfht.ml/2wr6uk9 . Probably best to have jandem, sfink, or literally anybody from the JS team to look at it to see if there's more work to be done there.

There's some jank where it looks like the compositor (along with the compositor sampling thread, so I can't be sure) was being starved: https://perfht.ml/2wria6r . That is causing the parent to block while it tries to synchronously send an APZ mouse event. I think bug 1366358 should help with this.

There's a big hang in the main thread of the parent process right at the end of the profile, but that looks to be in the guts of the OS when we're trying to peek a UI event, so I don't think there's much we can do about that.

	Bobby Chien
	Comment 7 • 7 years ago

tracking under quantum flow

	Ionut Budeanu
	Comment 8 • 7 years ago

Updating the Gecko Profile based on the latest benchmark measurements.

Name: Firefox
Version: 57.0a1
Windows 10 64 bit
buildID: 20170910220126

Gecko Profile: https://perfht.ml/2w14lfg

	Jet Villegas (inactive)
	Comment 9 • 7 years ago

(In reply to Mike Conley (:mconley) (:⚙️) from comment #4)
> mstange and I looked at this profile today in The Joy of Profiling Episode
> 8[1].
> 
> Most of the profile is JS bound. Folks from the JS team (perhaps sfink or
> jandem?) should look at this to see if there's anything new in here that's
> actionable.

ni? naveed to see if there's someone who can pick this up on JS team.

	Naveed Ihsanullah [:naveed]
	Comment 10 • 7 years ago

Steve please take a look. Is this something that still exist in FF57? If so should we address it in beta?

	Kannan Vijayan [:djvj]
	Comment 11 • 7 years ago

It seems like the remaining issues in this are simply slow JS.  Looking at the time-spent in the profile, the stack seems to primarily decompose into non-jitted script calls, likely due to the first-run nature of the test.

	Steve Fink [:sfink] [:s:]
	Comment 12 • 6 years ago

Yeah, I'm not seeing anything actionable here, nice as it would be to speed this up.