Hi Guys,
We currently have mysource matrix configured on a UNIX sun solaris 10 with postgresql, we have also loaded squid cache. Posgres has been performance tuned as recommended by Squiz. We have also loaded e-accelerator, and are currently load testing the site using WAPT as a testing tool.
When we execute the test with 5 - 20 virtual users postgres processes seem to hit between 15% - 20% CPU time. When this occurs the box is running at 100% CPU utilization. If the website is accessed while the above test is executing the home page and subsequent pages takes 10 - 15 seconds to load. The website is Cached and the HTTP headers are turned to on.
Could you please advise on what could be done to enhance performance.
It sounds like caching is not as active as expected, as after the first page load, Squid should be handling all the subsequent requests to the page, not Matrix. Have you configured your test suite to use the Squid proxy or is it hitting Matrix directly?
Also, there are a number of things that may be causing performance concerns, including inefficient designs, content nesting and other dynamic operations. The only way we can really advise you properly is to investigate your site+designs and make appropriate recommendations.
PostgreSQL on Solaris can also be a problem. Performance is nowhere near that of PostgreSQL on Linux for Matrix.
We are currently running the latest version of mysource matrix on solaris 10. Hardware is a sun server v245: 1x1.5ghx,2GBRAM, 2x146gb HDD.The database is postgresql version 8.1.3 with php version 4.4.7. We have also installed squid cache and php e-acellerator. The web server is apache. All installed on a standalone server. We are experiencing slow performance when running a WAPT web tool using 100% cpu processors, when running 20 virtual users.
Has anyone setup mysource on a simular platform? I would be interested in the performance stats of your website your installation set up or any information that may assist me
Is everything on the same box, i.e. Squid, Apache/Matrix and PostgreSQL? Can you confirm in the Squid logs that Squid is serving the pages and retrieving the content from cache?
Yes we tested this yesterday with the squid cache turned on and off the CPU utilisation was still at 100%. Postgres sessions on the box is varies between 15% to 25% cpu utilisation with about 10 postgres sessions running
Sounds a bit odd that having Squid in place doesn't speed anything up. Would suggest that Squid isn't actually caching anything.
Yes sounds odd to me. Read the http headers as pages are loaded (live hhtp headers for firefox or yslow plugins are useful). You should see expiry headers, and when you click around hardly anything should be loaded after the first page, only the body. Make sure everything is live and public read before you test. Try http://www.askbootshealth.com/ as an example of the sort of response you should be getting if properly tuned - its a fair bit of work but worth it.