I’ve been playing with the Authcache module the past couple of days. So far I’m very impressed.
I ran some Funkload tests with 1, 25, 50, and 100 concurrent clients to see how it does.
Pressflow 6.14.64 with external cache enabled:
http://staff.osuosl.org/~gchaix/test_simple-20091121T191202/#request-stats
Pressflow 6.14.64 with core cache replaced with Authcache:
http://staff.osuosl.org/~gchaix/test_simple-20091121T025003/#request-stats
Quite a difference at least in my initial tests. A roughly 50% increase in requests per second at 100 concurrent users, and a huge difference in the server load for the same test. The run queue during the external cache test spiked over 100 and a bunch of Apache threads stacked up. The run queue during the Authcache test stayed below 10 and the Apache threads disappeared so quickly they weren’t even visible in apachetop. The question is, why?
Authcache was configured to push the cache into my APC cache via Cacherouter for anonymous and authenticated users. Clearly it worked. The Pressflow external cache appeared to have little or no effect. That leads me to believe I need to take another look at my Varnish config, as it appears to not be caching things Pressflow is sending it.
Anyone else out there trying this? I’d love to compare notes/configs.
Have you tried tracing the requests with varnishlog? You can also use varnishstat to view real-time results of cache performance during the tests.
In any case, the Pressflow team is here to help get this working. A proper use of Varnish should blow away anything that requires handling most requests through Apache.
Thanks, David. I’m still just getting started with the testing. I totally agree that the Pressflow external cache should completely blow authcache out of the water. I’m certain it’s my suboptimal Varnish config. I’ll dig into the Varnish logs more tomorrow.