556198 - Investigate not using jemalloc in mozilla-runtime built for OOPP only

Status: RESOLVED FIXED

(Core Graveyard :: Plug-ins, defect)

Description

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

jemalloc might help plugin perf as it has browser and JS perf.  But Karl rightly points out that it might cause plugin perf degradation if plugins are tuned to the libc allocator.

Not really sure how to go about this other than finding some flash et. al. benchmarks and seeing how metrics change with jemalloc.

Comment 1

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Actually, mozilla-runtime is used for content processes, in which we'll always want jemalloc.  Updating bug title to reflect possibility of plugins being measurably faster with libc allocator.

Comment 2

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Build jemalloc into mozilla-runtime

Will look into plugin perf tomorrow.

Comment 3

[Karl Tomlinson (:karlt)]

(In reply to comment #1)
> Actually, mozilla-runtime is used for content processes, in which we'll always
> want jemalloc.

Good point.

> Updating bug title to reflect possibility of plugins being
> measurably faster with libc allocator.

More than perf, what I was thinking of with plugins is that there have been some bad interactions with jemalloc.  For browser and content processes, we can control (more of) the code and avoid bad interactions.  With plugins, we can't control (so much of) the code, and so it seemed safer to stick with libc.

However, AFAIK, we don't have any unresolved bad interactions between plugins and jemalloc.  As we are going to want jemalloc for content processes, perhaps we should just take the simple path and always use jemalloc in mozilla-runtime, assuming that we'll continue to use the same executable for plugin and content processes.

Comment 4

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Totally unscientific tests on the "industry standard" Bubblemark show

 flash w/ libc allocators:  stable ~102-104 fps
 flash w/ jemalloc:         stable ~101-103 fps

I don't really know of a good benchmark for plugin memory allocators, and if we're talking no detectable change on this one at least, seems fine to switch to jemalloc.

Comment 5

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

http://hg.mozilla.org/mozilla-central/rev/da3c19f5ac5e

No reason to backport this.