1731604 - 200.07 - 4.52% perf_reftest bloom-basic-2.html / tp5o + 1 more (Windows) regression on Thu September 16 2021

Status: RESOLVED FIXED

(Core :: DOM: Content Processes, defect, P2)

Description

[Andra Esanu]

Perfherder has detected a talos performance regression from push 55fa1d670b119502b16f8795990f9456e1bfcd8c. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
200%	perf_reftest bloom-basic-2.html	windows10-64-shippable-qr	e10s fission stylo webrender	1.28 -> 3.84
29%	cross_origin_pageload	windows10-64-shippable-qr	e10s fission stylo webrender	250.28 -> 322.84
5%	tp5o	windows10-64-shippable-qr	e10s fission stylo webrender	262.57 -> 274.43

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
66%	perf_reftest bloom-basic.html	windows10-64-shippable-qr	e10s fission stylo webrender	4.53 -> 1.53
11%	perf_reftest_singletons inline-style-cache-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	1,152.46 -> 1,028.38
7%	perf_reftest_singletons link-style-cache-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	475.15 -> 440.89
6%	perf_reftest_singletons link-style-cache-1.html	linux1804-64-shippable-qr	e10s fission stylo webrender	477.54 -> 450.31

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

[Chris Peterson [:cpeterson]]

same-origin loads are saved in the bfcache, which will affect when we process switch.

Nika says we should check if there is an existing process in the bfcache we should reuse when we swap BrowsingContextGroups.

Comment 2

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

== Change summary for alert #31362 (as of Sat, 18 Sep 2021 15:02:58 GMT) ==

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
24%	amazon dcf	windows10-64-shippable-qr	fission warm webrender	361.21 -> 446.21
22%	amazon ContentfulSpeedIndex	windows10-64-shippable-qr	fission warm webrender	398.25 -> 486.08
22%	amazon SpeedIndex	windows10-64-shippable-qr	fission warm webrender	396.83 -> 482.33
20%	amazon fcp	windows10-64-shippable-qr	fission warm webrender	330.92 -> 396.71
19%	amazon fnbpaint	windows10-64-shippable-qr	fission warm webrender	337.58 -> 402.38
18%	amazon PerceptualSpeedIndex	windows10-64-shippable-qr	fission warm webrender	404.50 -> 477.33
17%	bing-search FirstVisualChange	linux1804-64-shippable-qr	fission warm webrender	240.00 -> 280.00
16%	amazon FirstVisualChange	windows10-64-shippable-qr	fission warm webrender	374.79 -> 434.92
16%	bing-search ContentfulSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	261.67 -> 302.33
14%	bing-search loadtime	linux1804-64-shippable-qr	fission warm webrender	217.42 -> 248.71
...	...	...	...	...
10%	bing-search dcf	linux1804-64-shippable-qr	fission warm webrender	232.88 -> 255.08
7%	bing-search PerceptualSpeedIndex	linux1804-64-shippable-qr	fission warm webrender	517.50 -> 555.58
7%	instagram LastVisualChange	windows10-64-shippable-qr	fission warm webrender	844.54 -> 905.00
6%	twitter ContentfulSpeedIndex	macosx1015-64-shippable-qr	fission warm webrender	842.67 -> 895.92
5%	bing-search LastVisualChange	linux1804-64-shippable-qr	fission warm webrender	800.00 -> 840.00

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=31362

Comment 3

[Nika Layzell [:nika] (ni? for response)]

I'm optimistic that these regressions will also be fixed by the changes in bug 1731792.

Comment 4

[Chris Peterson [:cpeterson]]

(In reply to Nika Layzell [:nika] (ni? for response) from comment #3)

I'm optimistic that these regressions will also be fixed by the changes in bug 1731792.

@ Randell, do you think the improvements below from Nika's patch are good enough that we can resolve this bug as fixed?

Nika's patch improved perf_reftest bloom-basic-2.html, but the results are still noisy and not entirely back to the previous baseline:

https://treeherder.mozilla.org/perfherder/graphs?highlightAlerts=1&highlightChangelogData=1&series=autoland,3869358,1,1&timerange=7776000

Similar improvements on cross_origin_pageload, but not entirely back to the previous baseline:

https://treeherder.mozilla.org/perfherder/graphs?highlightAlerts=1&highlightChangelogData=1&series=autoland,3869400,1,1&timerange=2592000

tp5o, OTOH, looks fixed:

https://treeherder.mozilla.org/perfherder/graphs?timerange=2592000&series=autoland,3869735,1,1

Comment 5

[Chris Peterson [:cpeterson]]

Resolving as fixed because Randell says the new test results with Nika's fix for bug 1731792 are good enough.

Comment 6

[Chris Peterson [:cpeterson]]

Nika's fix for bug 1731792 landed in Nightly 94.