1667696 - CONTENT_FULL_PAINT_TIME/CONTENT_FRAME_TIME regression on 2020-07-28

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Matt Woodrow (:mattwoodrow)]

Looking at the dashboard it appears there was a regression on 2020-07-28 - https://metrics.mozilla.com/public/sguha/ds-283/dashboard.html

CONTENT_FULL_PAINT_TIME goes from 6.84 -> 8.01
CONTENT_FRAME_TIME foes from 0.10 -> 0.11

The regression range from the first nightly build on the 27th to the last on the 28th is this: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=798bdad605b985a71cd204a56bb816dea503d432&tochange=7b8c4f32cdde947c1b1f942905fbc5039b6d1c65

Given that this regresses CONTENT_FULL_PAINT_TIME and not CONTENT_PAINT_TIME or COMPOSITE_TIME, it seems likely that the regression is in display list deserialization/scene building.

Nothing in that range seems particularly likely, except maybe Bug 1648405, or Bug 1623792.

Nical, do you think it's possible that the switch to crossbeam could cause slowness here?

Comment 1

[Nicolas Silva [:nical]]

I would be surprised that the switch to crossbeam channel caused a regression on windows, it even improved out tscrollx score by 4~5% on linux and other projects have reported seeing improvements moving from std channels to crossbeam.

It should be easy enough to switch back to std channels on windows and see if it fixes the slowness.

Comment 2

[Glenn Watson [:gw]]

We decided to (temporarily?) revert the patch that adds the IS_IDENTITY flag as it's an easy patch to revert and see if it has any effect on that telemetry number.

Comment 3

[Nicolas Silva [:nical]]

Attachment: Bug 1667696 - Use standard rust channels on Windows. r=mattwoodrow

While we measuered somewhat surprisingly high performance improvements on linux when replacing standard channels with crossbeam ones on Linux, there has been a CONTENT_FRAME_TIME regression on Windows around the same time. In doubt, this patch makes us use standard channels on Windows to see if it fixes the regression. This patch will be reverted if it doesn't turn out restore the CONTENT_FRAME_TIME numbers.

Swgl needs to continue using crossbeam because it depends on select which doesn't exist in standard channels.

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/e6d7b596358d
Use standard rust channels on Windows. r=jrmuizel

Comment 5

[Bogdan Tara[:bogdan_tara | bogdant]]

https://hg.mozilla.org/mozilla-central/rev/e6d7b596358d

Comment 6

[Nicolas Silva [:nical]]

Let's keep it open at least until we have validated that not using crossbeam channels in Windows fixed the issue.

Comment 7

[Jim Mathies [:jimm]]

Looking good -
https://metrics.mozilla.com/public/sguha/ds-283/dashboard.html