1608423 - 0.26 - 0.28% Base Content JS (windows7-32, windows7-32-shippable) regression on push d389e3aa4522b2ecab85197572facdb50520ae36 (Thu January 9 2020)

Status: VERIFIED FIXED

(Core :: JavaScript: GC, defect)

Description

[Marian Raiciof [:marauder]]

We have detected an awsy regression from push:

https://hg.mozilla.org/integration/autoland/pushloghtml?changeset=d389e3aa4522b2ecab85197572facdb50520ae36

As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

0.28% Base Content JS windows7-32 opt 2,931,709.33 -> 2,939,930.67
0.28% Base Content JS windows7-32-shippable opt 2,931,722.67 -> 2,939,929.33
0.26% Base Content JS windows7-32-shippable opt 2,932,408.67 -> 2,939,920.00

You can find links to graphs and comparison views for each of the above tests at: https://treeherder.mozilla.org/perf.html#/alerts?id=24636

On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a treeherder page showing the jobs in a pushlog format.

To learn more about the regressing test(s), please see: https://wiki.mozilla.org/AWSY/Tests

Comment 1

[Jon Coppeard (:jonco)]

This was one of the potential consequences of bug 1424934. There are some things we can try to reduce this overhead however.

Comment 2

[Liz Henry (:lizzard) (relman/hg->git project)]

Doesn't look like a blocker for the release but happy to take any mitigations in 74 of course.

Comment 3

[Paul Bone [:pbone]]

(In reply to Jon Coppeard (:jonco) from comment #1)

This was one of the potential consequences of bug 1424934. There are some things we can try to reduce this overhead however.

I have an idea. use two bits per cell as we did before, but IF we want to do concurrent marking then allocate with jemalloc a separate mark bytemap with one byte per cell. This means that we only pay for the additional memory cost if we actually intend to do concurrent marking, which we may only do for foreground tabs anyway.

Comment 4

[Jon Coppeard (:jonco)]

(In reply to Paul Bone [:pbone] from comment #3)
Well, I don't want us to have to do an extra malloc when allocating an arena (or have to separate implementations of how the mark bits are stored).

Comment 5

[Jon Coppeard (:jonco)]

I expect this to go away now that bug 1424934 has been backed out.

Comment 6

[Ryan VanderMeulen [:RyanVM]]

NI Marian to verify.

Comment 7

[Marian Raiciof [:marauder]]

Yes, indeed there is an improvement on Jan 21st 4:45 AM for 5fda66dee91f6 - the back out.

https://treeherder.mozilla.org/perf.html#/graphs?highlightAlerts=1&selected=1959153,1031292127&series=autoland,1959153,1,4&timerange=1209600&zoom=1579619857499,1579631069314,2927712.044207353,2949046.789958895

I will close this ticket as fixed and the alert as backed out.

Thank you!