FS#51077 - [firefox] Use jemalloc

Attached to Project: Arch Linux
Opened by Constantine (Hi-Angel) - Friday, 30 September 2016, 12:03 GMT
Last edited by Jan de Groot (JGC) - Monday, 03 October 2016, 15:03 GMT
Task Type Feature Request
Category Packages: Extra
Status Closed
Assigned To Jan de Groot (JGC)
Evangelos Foutras (foutrelis)
Jan Alexander Steffens (heftig)
Architecture All
Severity Low
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Looking the "about:buildconfig", there's no "--enable-jemalloc" configure option, which Windows users report to have.

"jemalloc" is an implementation of memory allocation that reported to have a better performance (at least for firefox case).
This task depends upon

Closed by  Jan de Groot (JGC)
Monday, 03 October 2016, 15:03 GMT
Reason for closing:  Won't implement
Comment by Constantine (Hi-Angel) - Saturday, 01 October 2016, 05:33 GMT
Btw! I also notice, libc (alternatively it'd be libjmalloc) is statically linked in. Unless both the static library and Firefox are built with "-flto" (which isn't the case), there's no pros. I suggest to build this dynamically.
Comment by Constantine (Hi-Angel) - Saturday, 01 October 2016, 07:42 GMT
Btw! I also notice, libc (alternatively it'd be libjmalloc) is statically linked in. Unless both the static library and Firefox are built with "-flto" (which isn't the case), there's no pros. I suggest to build this dynamically.
Comment by Constantine (Hi-Angel) - Saturday, 01 October 2016, 07:44 GMT
Hmm, I'm sorry, I didn't send the duplicate. I just reloaded the page, Firefox for whatever reason asked if I want "To resend information" or "Cancel", which ended up with this.
Comment by Jan de Groot (JGC) - Sunday, 02 October 2016, 20:49 GMT
Upstream uses --enable-jemalloc in the common-opt mozconfigs for win32 and win64, but not for the linux builds.

I don't think it's wise to divert from upstream decisions on memory allocators here.
Comment by Constantine (Hi-Angel) - Sunday, 02 October 2016, 21:05 GMT
I see, you're wondering if that actually would give any performance ups.

Well, I actually came here from the comments of the article about Mesa switch to jemalloc¹, and for them it indeed gave a performance increase. I suppose, i.e. libc didn't switch to that algorithm because it may be benefits to specific workloads, but as Windows version did the switch long ago, Mozilla developers have probably found Firefox to be that kind of workload.

[1]: https://www.phoronix.com/scan.php?page=news_item&px=Mesa-RFC-Jemalloc
Comment by Jan de Groot (JGC) - Monday, 03 October 2016, 15:03 GMT
If you execute "strings /usr/bin/firefox | grep jemalloc" you'll find jemalloc all over the place, so probably linux builds use jemalloc by default.

I don't care about the performance stuff, it's about stability. All upstream tests on linux are done with the default settings, deviating from those will make our builds extremely untested and problems will be hard to debug.

Loading...