This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [patch] Make the mmap/brk threshold in malloc dynamic to improve performance
- From: Arjan van de Ven <arjan at linux dot intel dot com>
- To: Steve Munroe <sjmunroe at us dot ibm dot com>
- Cc: libc-alpha at sources dot redhat dot com, val_henson at linux dot intel dot com
- Date: Thu, 02 Mar 2006 22:08:51 +0100
- Subject: Re: [patch] Make the mmap/brk threshold in malloc dynamic to improve performance
- References: <OFCE81F4A3.5345E0DB-ON86257125.007315ED-86257125.0073AEC7@us.ibm.com>
Steve Munroe wrote:
libc-alpha-owner@sourceware.org wrote on 03/02/2006 01:16:56 PM:
> Hi,
>
> We've been investigating a performance issue that in the end turns out
> to be best fixed in glibc.
>
> The test case can best be described as a threaded application that has a
> pool of worker threads that, for a transaction, allocate a piece of
> memory (of around 1 to 8Mb, depending on the configuration), use that
> for the transaction and then free it again, and return to the pool.
>
A oprofile of the application before and after your proposed patch would
help demonstrate the problem and how the proposal improves the situation.
I have the kernel profiles to show it; I just need to organize them in a
presentable way ;)
Also you may want to look the existing bugzilla
http://sources.redhat.com/bugzilla/show_bug.cgi?id=1541 and see if the
problem you found was similar and if the test case attached to the this
bugzilla shows similar improvement.
I suspect it will from having looked at it on a high level