On Oct 18, 2007, at 1:10 AM, Mattias Jiderhamn wrote:
Btw, here is a more recent "comparison" on the same blog: http:// raibledesigns.com/rd/entry/comparing_open_source_application_servers Still just a single thread though, so still not a performance comparison. (And this time the author hints to be aware of that too) This time Resin (supposedly GPL) and Tomcat has about the same results (as do the others).
Single thread is not bad as a benchmark, actually. You'd generally want to do two separate runs, once with a single thread and once with lots and lots of threads. The single thread shows serial overhead, while the multi-thread shows parallel/locking overhead.
I don't know what the specific test is but it looks like a fairly reasonable result. It depends on how representative the application is, which is why you always need to benchmark with your own application.
The biggest difference in web-server performance will be the hello, world benchmark. Once you load the benchmark up with application code (and spring and hibernate/jpa and jsf and webservices and load- balancing overhead), the differences in the web-server become a smaller component of the total time. Unless, of course, your application is correctly setting its HTTP cache headers because that would test the performance of the proxy cache which is web-server dependent.
Once Resin's jpa, jsf, and webservices are complete and tuned, the benchmarks will become more important again because we'll have a bigger chunk of the total code under Resin control.
-- Scott
_______________________________________________ resin-interest mailing list [email protected] http://maillist.caucho.com/mailman/listinfo/resin-interest
_______________________________________________ resin-interest mailing list [email protected] http://maillist.caucho.com/mailman/listinfo/resin-interest
