1791547 - Enable balanced heap limits

Status: REOPENED

(Core :: JavaScript: GC, task, P1)

Description

[Jon Coppeard (:jonco)]

The last performance tests I did showed no regressions and it's the start of the cycle, so let's try enabling balanced heap limits everywhere. If this causes problems we can back it out straight away.

Note to sheriffs: this may cause performance changes.

Comment 1

[Jon Coppeard (:jonco)]

Attachment: Bug 1791547 - Enable balanced heap limits by default r?sfink

This is not enable for fuzzing builds or in the shell if the fuzzing safe
option is passed.

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/4c043a955561
Enable balanced heap limits by default r=sfink

Comment 3

[Marian-Vasile Laza]

https://hg.mozilla.org/mozilla-central/rev/4c043a955561

Comment 4

[Cristian Tuns]

Backout for causing Bug 1792394
Backout link: https://hg.mozilla.org/integration/autoland/rev/e8b8ae48f130fdf5a623a0d2b8bc2e0e8f976b44

Comment 5

[Sandor Molnar[:smolnar]]

Backout merged to central : https://hg.mozilla.org/mozilla-central/rev/4c043a955561

Comment 6

[Alexandru Ionescu (needinfo me) [:alexandrui]]

(In reply to Pulsebot from comment #2)

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/4c043a955561
Enable balanced heap limits by default r=sfink

== Change summary for alert #35516 (as of Mon, 26 Sep 2022 05:24:53 GMT) ==

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
4%	imdb loadtime	android-hw-a51-11-0-aarch64-shippable-qr	warm webrender	3,107.54 -> 3,241.12

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=35516

Comment 7

[Steve Fink [:sfink] [:s:]]

The landing of this showed a spike in matrix-react time, probably from doing another GC. May be useful to look at when tuning.

Comment 8

[Alexandru Ionescu (needinfo me) [:alexandrui]]

(In reply to Cristian Tuns from comment #4)

Backout for causing Bug 1792394
Backout link: https://hg.mozilla.org/integration/autoland/rev/e8b8ae48f130fdf5a623a0d2b8bc2e0e8f976b44

== Change summary for alert #35586 (as of Tue, 04 Oct 2022 03:23:44 GMT) ==

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
22%	welcome fcp	linux1804-64-shippable-qr	fission warm webrender	54.96 -> 67.29
22%	reddit-billgates-post-2.top LastVisualChange	linux1804-64-shippable-qr	cold fission webrender	716.67 -> 876.67
16%	welcome loadtime	linux1804-64-shippable-qr	fission warm webrender	47.19 -> 54.71
8%	youtube fcp	windows10-64-shippable-qr	fission warm webrender	180.67 -> 194.42
5%	facebook-nav.friends loadtime	linux1804-64-shippable-qr	cold fission webrender	966.17 -> 1,018.92

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
20%	youtube loadtime	windows10-64-shippable-qr	fission warm webrender	1,486.62 -> 1,196.46
17%	youtube SpeedIndex	windows10-64-shippable-qr	fission warm webrender	1,348.42 -> 1,123.33
16%	youtube PerceptualSpeedIndex	windows10-64-shippable-qr	fission warm webrender	1,110.08 -> 932.17
15%	youtube ContentfulSpeedIndex	windows10-64-shippable-qr	fission warm webrender	1,314.17 -> 1,111.33
14%	reddit-billgates-post-2.hot LastVisualChange	windows10-64-shippable-qr	cold fission webrender	788.27 -> 679.08
...	...	...	...	...
3%	linkedin PerceptualSpeedIndex	windows10-64-shippable-qr	fission warm webrender	875.62 -> 852.33

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=35586

Comment 9

[BugBot [:suhaib / :marco/ :calixte]]

There's a r+ patch which didn't land and no activity in this bug for 2 weeks.
:jonco, could you have a look please?
If you still have some work to do, you can add an action "Plan Changes" in Phabricator.
For more information, please visit auto_nag documentation.

Comment 10

[Jon Coppeard (:jonco)]

I'm not planning to land this just yet.

Comment 11

[Mayank Bansal]

With the extensive work done around SM in the time since this was last tried, the upcoming allocator work, focus on sp3 (and deprioritization of jetstream2) , it may be informative to check if this is still useful.
Maybe worth pushing to try?

Comment 12

[Steve Fink [:sfink] [:s:]]

I believe there is a current intention to rework this and try to land some form of it, or part of it at least. I think tcampbell is the one looking at it right now.