Gunnar Wolf schreef: > Milo van der Linden dijo [Thu, Jul 24, 2008 at 08:21:55PM +0200]: >> I have a virtual machine ready with the latest svn cherokee, lighttpd >> and apache to rerun the benchmarks. I am only waiting for the 0.8.0 to >> reach stable. Then I hope the new benchmarks will find their way to the >> website. I have automated the installation script for >> cherokee/apache/lighttpd. It can be re-run with every new release ;-) >> Hope that partially answers one of the questions. > > Virtual machines should be avoided when testing for performance - you > are incurring on the penalties of competing for resources _twice_ (on > each OS layer). If you want to give true statistics, do it as close as > possible to the metal!
The method I proposed before was using some old machines (aka 486/PII) and allowing a fixed amount of time of requests. The slower the machine the better. But there is one thing in VMs that are actually very cool. That is the amount of cpu cycles used. So instead of of 10 seconds of requests, you could actually do 50000 request, and check the CPU usage for each implementation. It is not about the maximum performance on a Dell1950, because a kernel can make a difference. It is how much performance is required to deliver peak performance. Stefan _______________________________________________ Cherokee mailing list [email protected] http://lists.octality.com/listinfo/cherokee
