[Ecls-list] Release candidate?

Matthew Mondor mm_lists at pulsar-zone.net
Thu May 31 04:13:00 UTC 2012


On Wed, 30 May 2012 23:22:41 +0200
Juan Jose Garcia-Ripoll <juanjose.garciaripoll at googlemail.com> wrote:

> With the latest fixes of today I think that ECL is ready for a new release.
> The most recent commit is
> 
> commit d545918cddedab3a5f1d5604a792159f2769c607
> Author: Juan Jose Garcia Ripoll <jjgarcia at users.sourceforge.net>
> Date:   Wed May 30 22:42:38 2012 +0200
> 
>     Additional fixes to build with GMP_TYPE=amd64
> 
> I would appreciate if you could test it and give me the ok.

I hadn't updated ECL in a few weeks at least, before tonight.  I
updated now, and unfortunately can see some performance degradation
with the HTTPd if using -n5000 with ab, rather than -n1000:

-n1000:
Requests per second:    2577.32 [#/sec] (mean)

-n5000:
Requests per second:    46.20 [#/sec] (mean)

Such behavior used to happen before, but that was before the new
locking implementation; at the time I reported a successful 30 days
uptime, I could run with -n10000 even without such issues.
Interestingly -n is not increasing concurrency (I used -n16 in all
tests), but it's the number of consecutive connections/requests.
I'm not sure if the problem is consing, gc-related and/or lock
contention/races related.  I didn't update my local boehm-gc and gmp
packages, so those are the same as when it worked better not that long
ago.

The four cores however all appear to be very busy during a run.
I didn't check yet the actual changes since I last updated, but could
possibly do that tomorrow.  Sorry for the annoying news...

Environment:

NetBSD-6/amd64 i5-2500/8GB (process seems to stop growing at ~93M
resident, so not a memory issue)

ECL from GIT HEAD,
boehm-gc-7.1nb4,
gmp-5.0.2nb1,
libffi-3.0.9nb1

Build:

export CFLAGS="-O2 -g"
export LDFLAGS='-g'
./configure --prefix=/usr/local/ecl --enable-unicode --enable-threads --with-__thread=no --enable-rpath --with-system-boehm=yes --with-system-gmp=yes --with-gmp-prefix=/usr/pkg --with-dffi=system

Thanks,
-- 
Matt




More information about the ecl-devel mailing list