800063 - Make GC_ALLOCATION_THRESHOLD dynamic

Status: RESOLVED FIXED

(Firefox OS Graveyard :: General, defect)

Description

[Gregor Wagner [:gwagner]]

In the desktop browser we only start GCing after the GC heap reached 30MB. We should make this dynamic and reduce it for mobile devices.

Comment 1

[Gregor Wagner [:gwagner]]

(In reply to Gregor Wagner [:gwagner] from comment #0)
> In the desktop browser we only start GCing after the GC heap reached 30MB.
> We should make this dynamic and reduce it for mobile devices.

This is the allocation based trigger. We sure can trigger GCs for other events.
But in the parent we have less (to none?) window-event based triggers.

Comment 2

[Justin Lebar (not reading bugmail)]

> But in the parent we have less (to none?) window-event based triggers.

We can definitely add them, though.  It would be a matter of bubbling the event up from BrowserElementChild to BrowserElementParent and then firing an observer.

Comment 3

[Andrew McCreight [:mccr8]]

I could even imagine doing a selective compartment GC that just works on the compartments in the parent that are associated with the child somehow. (Not knowing precisely how that all works...)

Comment 4

[Gregor Wagner [:gwagner]]

(In reply to Andrew McCreight [:mccr8] from comment #3)
> I could even imagine doing a selective compartment GC that just works on the
> compartments in the parent that are associated with the child somehow. (Not
> knowing precisely how that all works...)

I guess that's what we are currently doing. In the parent I mostly see malloc and alloc based triggers for one compartment.

Hm a huge win could be a global GC that is connected to the idle API.

Comment 5

[Justin Lebar (not reading bugmail)]

> Hm a huge win could be a global GC that is connected to the idle API.

Along the same lines, I just filed bug 800166.

Comment 6

[Andrew Overholt [:overholt]]

Marking blocking as it seems to be a big enough win to justify.

Comment 7

[Gregor Wagner [:gwagner]]

Attachment: patch

Comment 8

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Comment on attachment 675712 [details] [diff] [review]
patch

Review of attachment 675712 [details] [diff] [review]:
-----------------------------------------------------------------

One problem I see here is that incremental GC isn't used for the allocation-based triggers. But it seems like you guys are mostly worried about memory usage, so maybe it's better to do it this way.

Comment 9

[Gregor Wagner [:gwagner]]

You are right, the current focus is memory usage. We can still tweak it if it causes too many GCs.

https://hg.mozilla.org/integration/mozilla-inbound/rev/a32865b52f01

Comment 10

[Gregor Wagner [:gwagner]]

Followup:
https://hg.mozilla.org/integration/mozilla-inbound/rev/647d6f4cd15d

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a32865b52f01
https://hg.mozilla.org/mozilla-central/rev/647d6f4cd15d

Comment 12

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/c25644a55f65
https://hg.mozilla.org/releases/mozilla-aurora/rev/7af481b6c252