This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: malloc: Trim unused arenas on thread exit
- From: DJ Delorie <dj at redhat dot com>
- To: Florian Weimer <fweimer at redhat dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Mon, 13 Nov 2017 16:57:22 -0500
- Subject: Re: malloc: Trim unused arenas on thread exit
- Authentication-results: sourceware.org; auth=none
Florian Weimer <fweimer@redhat.com> writes:
> I think trimming unused arenas as much as possible is a good heuristics
> to minimize RSS, so getting this to work might be worthwhile.
I ran this against many of my real-world workloads and did not see a
significant (i.e. obvious) reduction in actual RSS. Some tests showed a
minor (i.e. untrustworthy, I didn't run enough passes to reduce the
noise below any measured change) *increase* in RSS, as well as an
expected (but still untrustworthy) increase in cycles required.
Now that we have the tools to *measure* what our malloc changes cause, I
think it's prudent that we start asking for benchmarks (and workloads)
that back up any proposed performance/rss change.