1675035 - 3.24 - 11.35% bbc / bing-search-restaurants / bing-search-restaurants / booking PerceptualSpeedIndex / facebook LastVisualChange / instagram ContentfulSpeedIndex (android-hw-g5-7-0-arm7-api-16-shippable, android-hw-p2-8-0-android-aarch64-shippable) regr

Status: RESOLVED INCOMPLETE

(GeckoView :: General, defect, P2)

Description

[Florin Strugariu [:Bebe]]

Perfherder has detected a browsertime performance regression from push 4ecb44816920e876206d556cf7fa0d766dcacf09. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
11%	instagram	ContentfulSpeedIndex	android-hw-p2-8-0-android-aarch64-shippable	warm	485.50 -> 540.58
5%	bbc	PerceptualSpeedIndex	android-hw-p2-8-0-android-aarch64-shippable	warm	636.17 -> 667.92
5%	bing-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable	cold	1,651.50 -> 1,730.67
5%	bbc	SpeedIndex	android-hw-p2-8-0-android-aarch64-shippable	warm	590.17 -> 618.08
5%	bing-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable	warm	1,310.08 -> 1,371.17
5%	booking	PerceptualSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	warm	792.04 -> 828.75
4%	bbc	LastVisualChange	android-hw-p2-8-0-android-aarch64-shippable	warm	1,110.92 -> 1,160.50
4%	bing-search-restaurants	PerceptualSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	warm	986.33 -> 1,029.42
4%	bing-search-restaurants	PerceptualSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	cold	1,293.33 -> 1,346.50
4%	bing-search-restaurants	SpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	warm	787.92 -> 820.17
4%	bing-search-restaurants	ContentfulSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	warm	521.75 -> 542.33
4%	bbc	LastVisualChange	android-hw-p2-8-0-android-aarch64-shippable	warm	1,114.25 -> 1,157.67
3%	bing-search-restaurants	SpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable	cold	1,073.96 -> 1,111.17
3%	facebook	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable	warm	1,157.17 -> 1,194.67

Improvements:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
18%	allrecipes	ContentfulSpeedIndex	android-hw-p2-8-0-android-aarch64-shippable	warm	584.58 -> 478.92
9%	allrecipes	ContentfulSpeedIndex	android-hw-p2-8-0-android-aarch64-shippable	cold	1,363.12 -> 1,236.50

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) will be backed out in accordance with our regression policy.

For more information on performance sheriffing please see our FAQ.

Comment 1

[Hiroyuki Ikezoe (:hiro)]

So the cold case regressions are somewhat expected because bug 1500644 changed the browser (GeckoView example) structure, it's now bit complicated than before so it takes more time to create a browser window.

As for the warm regression cases, I am curios how bug 1500644 affects the regressions. Indeed, with the dynamic toolbar, some layout related metrics are different from the case without the dynamic toolbar, so it's maybe possible that dynamic toolbar related code may have some amount of performance impacts. That said, the dynamic toolbar related code has been there and Fenix has been using it, so this regression will be WONTFIX.

Anyway, I am curious how dynamic toolbar affects our layout performance, but unfortunately as of now we can't take profiles for browsertime tests on GeckoView due to bug 1675170 and bug 1675157 (or more).

Comment 2

[Ryan VanderMeulen [:RyanVM]]

These tests appear to have stopped running around mid-November. Is there a different test suite we should be looking at for this now?

Comment 3

[Dave Hunt [:davehunt] [he/him] ⌚BST]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #2)

These tests appear to have stopped running around mid-November. Is there a different test suite we should be looking at for this now?

We enabled WebRender, and then disabled conditioned profiles, causing the test signatures to change. Here's a graph showing the transitions for one of the affected tests. I believe it's still possible to investigate this regression, perhaps using a push to try with/without the suspected patch backed out and comparing results in Perfherder.