Posted by simg on August 16, 2012 at 7:35am
Has anyone got any experience of running Drupal (7) with http://quercus.caucho.com (compiles PHP into Java and apparently gives a 4x performance improvement).
There are some simple instructions on how to run Drupal 6 on "Resin 3.x". In the very near future I'm going to have a go with D7 on Resin 4.x and will let you know.
Comments
I'd be interested in hearing
I'd be interested in hearing your results with this, even if they are negative in that you don't get it running. Please follow up here if you get around to this.
I have tried and it
I have tried and it completely fails to mimc the reference counting feature of PHP. Drupal 7 relies on that in drupal_static(). Unfortunately, I think this is a dead end.
APC > Caucho
In my experience APC provides a better performance boost (doing basically the same thing) and is very well supported.
That said, Facebook is allegedly working on a similar project that would do a hip-hop style compiled binary, which would present a lot of efficiency gains.
http://www.chapterthree.com | http://www.getpantheon.com | http://www.outlandishjosh.com
Yes, I've tried the Hiphop
Yes, I've tried the Hiphop Virtual Machine. (as opposed to the HipHop PHP -> C compiler)
Couldn't get Drupal working with it (didn't try very hard) but some simple PHP benchmarks I ran suggested that that the performance improvement of HHVM (over APC) was ~30%
Facebook devs say they have the first 90% of HHVM complete, now they have to work on the second 90% -> http://www.facebook.com/note.php?note_id=10150415177928920
I'm curious how Pantheon handles performance for logged in users ? My current understanding (and I'm happy to be proven wrong) is that Drupal is (very) CPU bound in the themeing layer for logged in users ?
Use a set of modules like Views / Context / Display Suite / Panels / Rules and it appears you have a choice between slow dynamic page generation (350 - 1000+ ms) or static caching of page content ?
My question about Drupal performance expectations (http://groups.drupal.org/node/242798) never really got a satisfactory answer (no offence to those people who did response). I don't need help making my Drupal installations faster, I just need to know what other people have achieved so that I know what I'm aiming at. I suspect you would be a great person to answer this question :)
I have tested running a
I have tested running a fairly complex D6 application on it and it functions. Haven't done any performance comparisons, but I would doubt that it is more than 1.5x performance improvement (that is about all hphp can squeeze out on a properly optimized stack). At that, you're only looking at improving the performance of the PHP code. SQL queries, external services, etc are the typical offenders for poor performance and you won't see performance gains there.
Where I see the value of Quercus is native integration with Java libraries and the ability to run on enterprise platforms built for running Java applications and Java technologies (JMS, JCache, etc).
Similar to APC
From benchmarks that were done a few years ago, I remember that performance was noticeably faster than plain PHP, but comparable to PHP when APC is used, so not that much of a boost.
Drupal performance tuning, development, customization and consulting: 2bits.com, Inc..
Personal blog: Baheyeldin.com.
Thanks. Just re-read this
Thanks. Just re-read this article http://www.workhabit.com/labs/resin-backed-php-drives-4x-performance-imp... which makes the 4x improvement claim.
Very disingenuous to claim a 4x improvement over stock PHP with no op-code cache.
Many PHP speed comparisons aren't clearly tested
Just to second what @kbahey said, a lot of the PHP speedup schemes that are posted on the internet don't look so good compared to PHP with the APC cache. Many times it is not even clear if their comparisons are against plain PHP or PHP with the APC cache.
I think we depend on the APC cache a lot, it is under-appreciated for how good it is. It also has very few contributors; it would be good if some of the Drupal community started participating there.
I wonder if some of the tricks that the Java JIT / JVM opptimizers have done in the last decade could be applied to APC and speed us up ?