This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [RFC] Lock elision "test suite", first draft
- From: Dominik Vogt <vogt at linux dot vnet dot ibm dot com>
- To: libc-alpha at sourceware dot org
- Date: Fri, 9 Aug 2013 10:18:46 +0200
- Subject: Re: [RFC] Lock elision "test suite", first draft
- References: <20130806093649 dot GA23477 at linux dot vnet dot ibm dot com> <20130807120723 dot GA7829 at linux dot vnet dot ibm dot com> <20130807122010 dot GB5794 at domone dot kolej dot mff dot cuni dot cz>
- Reply-to: libc-alpha at sourceware dot org
On Wed, Aug 07, 2013 at 02:20:10PM +0200, OndÅej BÃlka wrote:
> On Wed, Aug 07, 2013 at 02:07:23PM +0200, Dominik Vogt wrote:
> > Test results on z/Architecture (zEC12)
> > --------------------------------------
> >
> > Test Setup:
> >
> > An lpar with eight CPUs on a ZEC12. 8 GiB RAM.
> >
> > Test execution:
> >
> > Each test case was run ten times in a loop using an unpatched
> > glibc-2.15 (Ubuntu standard) ("old glibc"), a new glibc-2.18
> > with elision disabled ("elision n/a"), and a new glibc-2.18
> > with elision enabled ("elision on").
> snip
> > ./ets-t3
> > =========
> >
> > old glibc:
> > ----------
> > pct_best : 100.00...%
> > pct_mindev: 4.57...%
> > pct_maxdev: 5.43...%
> > pct_stddev: 3.61...%
> >
> > elision n/a:
> > ------------
> > pct_best : 96.79...%
> > pct_mindev: 9.82...%
> > pct_maxdev: 9.42...%
> > pct_stddev: 5.71...%
> >
> > elision on:
> > -----------
> > pct_best 593.48...%
> > pct_mindev: 5.73...%
> > pct_maxdev: 4.09...%
> > pct_stddev: 3.43...%
> >
> Do you know why elision is 6 times slower on this test?
The throughput is six times of the throughput without elision.
The percentages of pct_best are relative performance compared with
glibc-2.15.
Vice versa, an explanation is needed for the performance of the
tests 1, 2, 4 and 5. I'm working on that and will post individual
analyses of the tests soon.
Ciao
Dominik ^_^ ^_^
--
Dominik Vogt
IBM Germany