From: Eric Wong <normalperson@...>
Date: 2018-05-23T02:57:39+00:00
Subject: [ruby-core:87230] Re: [Ruby trunk Feature#14718] Use jemalloc by default?

dennisb55@hotmail.com wrote:
> normalperson (Eric Wong) wrote:
> >  Reading some glibc mailing list (libc-alpha) posts last year,
> >  I remember excessive RSS usage is one of the areas they're
> >  tackling due to user complaints.

> Interesting, do you have links?

Wasn't even last year.
https://marc.info/?i=a9367dd0-f130-a23c-df7b-14d50ed10cfa@redhat.com
(or https://public-inbox.org/libc-alpha/a9367dd0-f130-a23c-df7b-14d50ed10cfa@redhat.com/ )

	A high-level concrete problem today with glibc's malloc, and the
	only problem be reported by our users is that it consumes too
	much RSS. Solving that problem in abstract is what we should be
	looking at.
	- Carlos O'Donell

> >  I should also note the newish glibc 2.26 has the thread-caching
> >  malloc as a build-time option, so it should reduce contention on
> >  arenas and allow using fewer arenas.

> I wonder what the fragmentation characteristics will be?

*shrug* haven't gotten around to testing

> The arena count defaults haven't changed have they? (8 times core count for x64).

Doesn't seem to have changed:

	git clone git://sourceware.org/git/glibc.git &&
		git -C glibc grep NARENAS_FROM_NCORES

Unsubscribe: <mailto:ruby-core-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>