[ruby-core:62912] [ruby-trunk - Feature #9113] Ship Ruby for Linux with jemalloc out-of-the-box

Issue #9113 has been updated by Motohiro KOSAKI.


 On Sat, May 24, 2014 at 9:48 PM,  <[e-mail]> wrote:
 > Issue #9113 has been updated by Eric Wong.
 >
 >
 >  KOSAKI Motohiro <[e-mail]> wrote:
 >  > Do we have only one benchmark provided by Sam? I don't think it is
 >  > enough much comparison to make decision.
 >
 >  Empirical evidence on my 32-bit yahns server after running several days
 >  shows ~40M RSS w/ jemalloc 3.0/3.6 vs 60-80M RSS for eglibc malloc on
 >  Debian stable.  I'll work on gathering more data for other systems.
 
 So, I'd suggest two phase action.
 
 1. Commit your patch, but disable by default.
 2. Gather more use-case and performance data.
     Note: They should be reproducible. We need measure them again and
 again when libs updated.
 3. Change the default when the community is convinced the benefit.

----------------------------------------
Feature #9113: Ship Ruby for Linux with jemalloc out-of-the-box
https://bugs.ruby-lang.org/issues/9113#change-47017

* Author: Sam Saffron
* Status: Feedback
* Priority: Normal
* Assignee: 
* Category: build
* Target version: 
----------------------------------------
libc's malloc is a problem, it fragments badly meaning forks share less memory and is slow compared to tcmalloc or jemalloc. 

both jemalloc and tcmalloc are heavily battle tested and stable. 

2 years ago redis picked up the jemalloc dependency see: http://oldblog.antirez.com/post/everything-about-redis-24.html 

To quote antirez:
``
But an allocator is a serious thing. Since we introduced the specially encoded data types Redis started suffering from fragmentation. We tried different things to fix the problem, but basically the Linux default allocator in glibc sucks really, really hard. 
``

--- 

I recently benched Discourse with tcmalloc / jemalloc and default and noticed 2 very important thing: 

median request time reduce by up to 10% (under both)
PSS (proportional share size) is reduced by 10% under jemalloc and 8% under tcmalloc.

We can always use LD_PRELOAD to yank these in, but my concern is that standard distributions are using a far from optimal memory allocator. It would be awesome if the build, out-of-the-box, just checked if it was on Linux  (eg: https://github.com/antirez/redis/blob/unstable/src/Makefile#L30-L34 ) and then used jemalloc instead. 

---Files--------------------------------
0001-configure.in-add-with-jemalloc-option.patch (1.29 KB)


-- 
https://bugs.ruby-lang.org/