1848354 - 10.43 - 5.31% wikipedia loadtime / wikipedia fcp + 3 more (OSX) regression on Mon August 7 2023

Status: NEW

(Core :: Internationalization, defect, P1)

Description

[Alex Finder]

Perfherder has detected a browsertime performance regression from push d6760db62dc1cc0831c09c42f086cf57d1c2bf48. 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)	Performance Profiles
10%	wikipedia loadtime	macosx1015-64-shippable-qr	bytecode-cached fission warm webrender	338.71 -> 374.05	Before/After
9%	wikipedia loadtime	macosx1015-64-shippable-qr	fission warm webrender	358.40 -> 392.18	Before/After
9%	wikipedia loadtime	macosx1015-64-shippable-qr	fission warm webrender	358.75 -> 390.29	Before/After
6%	wikipedia fcp	macosx1015-64-shippable-qr	cold fission webrender	525.81 -> 556.91	Before/After
5%	wikipedia fcp	macosx1015-64-shippable-qr	bytecode-cached cold fission webrender	527.87 -> 555.89	Before/After

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) may be backed out in accordance with our regression policy.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

For more information on performance sheriffing please see our FAQ.

Comment 1

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

Set release status flags based on info from the regressing bug 1719535

Comment 2

[Donal Meehan [:dmeehan]]

Tracking + since it's large change on a significant site

Comment 3

[Makoto Kato [:m_kato]]

This change is Nightly only now. No effect for beta and release.

I am interesting why macOS only. (Windows and Linux are no alert). LSTM?

Comment 4

[Donal Meehan [:dmeehan]]

Thanks, removing tracking for 118

Comment 5

[Makoto Kato [:m_kato]]

Looking the profiling data and debug internal again.

• Gecko side will pass word when finding a space. So it means that wikipedia's long page will call icu_segmenter a lot. (over 10,000).
• macOS's heap allocator seems to be slower than other platform. So it causes macOS only regression.
• We can optimize unbreakable path for AL, NU, IS and QU when is ASCII. This can reduce allocation number. And we should cleanup nsLineBreaker's fast path. (It doesn't match with UAX14). But it may not be enough.
• Create stack only FFI for Line breaker may resolve this, I need to try it.

Comment 6

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

Set release status flags based on info from the regressing bug 1719535

Comment 7

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

The severity field is not set for this bug.
:m_kato, could you have a look please?

For more information, please visit BugBot documentation.

Comment 8

[Maire Reavy [:mreavy]]

I'd really love this fixed, but this bug is an S3, not an S2. (S2 is a major bug that would cause a user to leave Firefox; S3 is a normal bug that we'd like, even love, to fix as soon as we have time.) Please chime in and needinfo me if you think this assessment is wrong. Thanks!

Comment 9

[Makoto Kato [:m_kato]]

I have added some improvements for segmenter code, then wikipedia fcp opt cold fission webrender of macOS is 524.1 and wikipedia loadtime opt fission warm webrender is 375. I will still need to watch perf data.