hi!
Here are a first result set using 5.5 and O+.
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
We did not update the templates for the report, please read the table as:
- No Cache:
. 5.5 VC11 PGO vs 5.4 VC9 PGO - Wincache
. 5.4 PGO + Wincache vs 5.5 PGO + O+
TS mode is totally broken using O+, so ignore this column.
Mediawiki is slower because it does not support yet O+ for usercache,
along other things. Please keep in mind that these tests are only the
performance related tests. Unit tests using all apps and phpts are
still running.
Further tests are running for:
- 5.4 PGO + APC vs 5.4 PGO + Wincache vs 5.4 PGO + o+
- 5.3 PGO + APC vs 5.3 PGO + Wincache vs 5.3 PGO + o+
- 5.5 PGO + APC vs 5.5 PGO + Wincache vs 5.5 PGO + o+
I will post the results as soon as they are available.
I would also suggest to publish releases as soon as possible on
pecl.php.net to increase the user base.
Thanks to Steve Zarkos and Matt Ficken for their work :)
Cheers,
Pierre
@pierrejoye
Hi Pierre,
Pierre Joye in php.internals (Fri, 15 Feb 2013 07:21:00 +0100):
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
We did not update the templates for the report, please read the table as:
- No Cache:
. 5.5 VC11 PGO vs 5.4 VC9 PGO- Wincache
. 5.4 PGO + Wincache vs 5.5 PGO + O+TS mode is totally broken using O+, so ignore this column.
Is that the column 5.5.0devvc11 under APC -igbinary? I have a Drupal7
site running on my devserver (W2K8, Apache 2.4.3, PHP 5.3.21 TS as
mod_fcgid) and do not get the idea it is running at one third (21.9 /
69.4) of the speed of the same site without any cache at all.
Do you have an idea what goes wrong?
Mediawiki is slower because it does not support yet O+ for usercache,
along other things.
Is the same the case with Symfony?
Jan
Hi Pierre,
Pierre Joye in php.internals (Fri, 15 Feb 2013 07:21:00 +0100):
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
We did not update the templates for the report, please read the table as:
- No Cache:
. 5.5 VC11 PGO vs 5.4 VC9 PGO- Wincache
. 5.4 PGO + Wincache vs 5.5 PGO + O+TS mode is totally broken using O+, so ignore this column.
Is that the column 5.5.0devvc11 under APC -igbinary? I have a Drupal7
site running on my devserver (W2K8, Apache 2.4.3, PHP 5.3.21 TS as
mod_fcgid) and do not get the idea it is running at one third (21.9 /
69.4) of the speed of the same site without any cache at all.
No, but O+ is broken right now in ZTS, it makes no sense to do further
testing before it works for more than 0.5 request ;)
But running TS builds in FCGI reduces drastically the issues due to
thread safety bugs, while it makes no sense to use them, fcgi and the
likes should rely on NTS builds. ZTS will do all kind of locks where
it is absolutely not required, that explains at least 15-20% of the
perf drop, virtual paths cover the rest.
Do you have an idea what goes wrong?
Not yet, Dmitry told me that he will work on it today.
Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
Cheers,
Pierre
@pierrejoye
Pierre Joye in php.internals (Fri, 15 Feb 2013 09:54:03 +0100):
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
Is that the column 5.5.0devvc11 under APC -igbinary? I have a Drupal7
site running on my devserver (W2K8, Apache 2.4.3, PHP 5.3.21 TS as
mod_fcgid) and do not get the idea it is running at one third (21.9 /
69.4) of the speed of the same site without any cache at all.No, but O+ is broken right now in ZTS, it makes no sense to do further
testing before it works for more than 0.5 request ;)But running TS builds in FCGI reduces drastically the issues due to
thread safety bugs, while it makes no sense to use them, fcgi and the
likes should rely on NTS builds. ZTS will do all kind of locks where
it is absolutely not required, that explains at least 15-20% of the
perf drop, virtual paths cover the rest.
I will try to compare the same site side-by-side with different configs.
The dev-server has 4 IPv4=adresses which gives a lot of config options.
Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
Is the performance drop for Symfony in the Wincache test due to the fact
that wincache is optimized for IIS or are there other reasons?
Jan
Pierre Joye in php.internals (Fri, 15 Feb 2013 09:54:03 +0100):
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
Is that the column 5.5.0devvc11 under APC -igbinary? I have a Drupal7
site running on my devserver (W2K8, Apache 2.4.3, PHP 5.3.21 TS as
mod_fcgid) and do not get the idea it is running at one third (21.9 /
69.4) of the speed of the same site without any cache at all.No, but O+ is broken right now in ZTS, it makes no sense to do further
testing before it works for more than 0.5 request ;)But running TS builds in FCGI reduces drastically the issues due to
thread safety bugs, while it makes no sense to use them, fcgi and the
likes should rely on NTS builds. ZTS will do all kind of locks where
it is absolutely not required, that explains at least 15-20% of the
perf drop, virtual paths cover the rest.I will try to compare the same site side-by-side with different configs.
The dev-server has 4 IPv4=adresses which gives a lot of config options.Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
Is the performance drop for Symfony in the Wincache test due to the fact
that wincache is optimized for IIS or are there other reasons?
Did not analyze it yet.
WinCache is not IIS optimized but FastCGI (read: process based). Also
we did not enabled file IO cache, which could bring it up in line.
But as neither APC or O+ have it, it cannot be compared.
Cheers,
Pierre
@pierrejoye
I assume that something is functionally wrong on Symfony/O+:
Symfony, IIS, cache, PHP5.5.0devvc11: 2195
Symfony, IIS, cache, PHP5.5.0devvc11: 1919
Symfony, IIS, cache, PHP5.5.0devvc11: 2127
There was a previous report that Twig did not work at all with
optimizations enabled.
Damien
Pierre Joye in php.internals (Fri, 15 Feb 2013 09:54:03 +0100):
http://windows.php.net/downloads/snaps/ostc/pftt/perf/results-20130213-5.4.11-5.5.0devvc11.html
Is that the column 5.5.0devvc11 under APC -igbinary? I have a Drupal7
site running on my devserver (W2K8, Apache 2.4.3, PHP 5.3.21 TS as
mod_fcgid) and do not get the idea it is running at one third (21.9 /
69.4) of the speed of the same site without any cache at all.No, but O+ is broken right now in ZTS, it makes no sense to do further
testing before it works for more than 0.5 request ;)But running TS builds in FCGI reduces drastically the issues due to
thread safety bugs, while it makes no sense to use them, fcgi and the
likes should rely on NTS builds. ZTS will do all kind of locks where
it is absolutely not required, that explains at least 15-20% of the
perf drop, virtual paths cover the rest.I will try to compare the same site side-by-side with different configs.
The dev-server has 4 IPv4=adresses which gives a lot of config options.Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
Is the performance drop for Symfony in the Wincache test due to the fact
that wincache is optimized for IIS or are there other reasons?Did not analyze it yet.
WinCache is not IIS optimized but FastCGI (read: process based). Also
we did not enabled file IO cache, which could bring it up in line.
But as neither APC or O+ have it, it cannot be compared.Cheers,
Pierre
@pierrejoye
Hello Damien,
-----Original Message-----
From: Damien Tournoud [mailto:damz@damz.org]I assume that something is functionally wrong on Symfony/O+:
Symfony, IIS, cache, PHP5.5.0devvc11: 2195 Symfony, IIS, cache,
PHP5.5.0devvc11: 1919 Symfony, IIS, cache, PHP5.5.0devvc11: 2127
Yes, that is correct. We've since corrected the issue for symfony and ran another test with better results. You can always find our test results at the following page (organized by date, top to bottom): http://windows.php.net/downloads/snaps/ostc/pftt/perf/
We'll be running these tests constantly. The next run should also include fixes for ZTS and other things, so please feel free to keep an eye on the latest results and let me know if you have questions.
Thanks!
Stephen A. Zarkos
Microsoft
Open Source Technology Center
Office: 425-704-0422
stephen.zarkos@microsoft.com
Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
As far as I understand, O+ doesn't have any user caching support, right?
So the comparison needs to be with user caching disabled in Mediawiki
when running with APC or Wincache.
Damien
Mediawiki is slower because it does not support yet O+ for usercache,
along other things.Is the same the case with Symfony?
no, MediaWiki caching modules only support APC or Wincache so far.
As far as I understand, O+ doesn't have any user caching support, right?
So the comparison needs to be with user caching disabled in Mediawiki
when running with APC or Wincache.
Hi Damien,
No, O+ doesn't have user level caching, it's only for opcode caching. It
was discussed that APC should continue to provide user level caching.
Damien
no, MediaWiki caching modules only support APC or Wincache so far.
As far as I understand, O+ doesn't have any user caching support, right?So the comparison needs to be with user caching disabled in Mediawiki
when running with APC or Wincache.
Hi Damien,No, O+ doesn't have user level caching, it's only for opcode caching. It
was discussed that APC should continue to provide user level caching.
Does that mean that MediaWiki was using O+ for opcode cache and APC
for user level cache?
The best way to compare them would be to test both opcode caches with
a memcached server. No interaction between extensions, and no difference
in the connections to access the caching store. Moreover, if you are using
a production setup with multiple servers, you need to use memcached.