1416180 - Non-incremental GC spends a lot of time waiting for background threads to finish

Status: NEW

(Core :: JavaScript: GC, enhancement, P3)

Description

[Jan de Mooij [:jandem]]

Noticed this on bug 1391710 - we allocate a lot so we switch to non-incremental GCs. This involves waiting for background threads to finish and this time shows up in profiles as being pretty significant.

It would be nice if we could either stay incremental or optimized the waiting part somehow.

Comment 1

[Jason Orendorff [:jorendorff]]

Paul, what do you think?

Comment 2

[Paul Bone [:pbone]]

We generally want to make all GCs incremental.  This is gradual and we're attaching the most common causes on non-incremental GCs first.  It depends exactly on what's happening in this case (I'll look next week).  Jonco recently changed the behaviour with malloc memory thresholds.  One of his goals was to reduce the likelyhood of a non-incremental GC due to a malloc memory threshold.  However the accounting for malloc memory was also changed and we'd like to know if that bug caused a regression in this test case.

I'm not sure about waiting for background tasks and what can be done.  I suspect something can be done about background sweeping but I havn't investigated further.

I'd prefer to work on doing fewer non-incremental GCs.  But I'll take a look at the profile.

I've also filed bug 1416541 for one potential improvment.

Comment 3

[Firefox Bug Husbandry Bot]

https://wiki.mozilla.org/Bug_Triage/Projects/Bug_Handling/Bug_Husbandry#Move_fix-optionals