In my organization we are using IIS7 in front of
Tomcat 6.0.26. Now, before I get lots of responses that say "Don't use IIS" or "Use Apache" or "Use Tomcat Standalone," trust me, if I could follow that advice I would. IIS is specifically required for our organization and using any other configuration is not an option.
We are doing some load
testing on an 8-core 3Ghz, 8GB RAM preproduction server and noticed that, with 75 simulated simultaneous users, we were seeing an average response time of 830ms for our most commonly-accessed static content (CSS, JavaScript and skin image files used on every page). This seemed a little too high to us, so we decided to have IIS serve our static content directly and bypass Tomcat. We followed all the instructions in the Tomcat documentation and DID get the configuration to work. However, average response times for the same subset of static content
increased from 830ms to 2950ms, which is extreme and surprising. All of these results are repeatable over and over again.
All conventional wisdom I could find online suggested that the opposite would happen, and my own experience as a programmer tells me that it should. We have checked all IIS performance options and everything is set to more-than adequate levels for the simulated load. Our
JSP and
Servlet response times did not change, and are much lower than 2950ms, so it seems unlikely (though not impossible) that some performance option in IIS could be causing this issue (seeing as how all the traffic is going through IIS one way or another).
Does anybody know what could be the cause here? We've spend days on this problem and we're not getting anywhere (and by we I mean myself a programmer, my manager and an IIS/server administrator). We just don't know what else to try.
Thanks,
Nick
