Hi Cameron:
Thanks for your continuing investigation and comments. And thanks for volunteering to help us test patches.
I just want to summarize what we think you're saying:
a) With your JMeter test, you can run to a higher throughput with clusterjsp than you can with GlassFish. Is that correct?
b) you're willing to take and try out these same tests with one or maybe a couple of test patch jars we will send you. Is that correct?
If so - again thanks. Jan and I are busy today teaching some training but perhaps by tomorrow we can send you a patch - maybe later today but no promises.
regards
It has been awhile since we've spent much time worrying about frames, but I wanted to ask you about one other thing. I think when you are using frames, they end up being processed separately (and asynchronously) as separate threads on the web container. I think that includes when they send back responses to the client too. So when you say JMeter client is "waiting for a response" before sending another request, I think you are creating a situation where many mini-requests and responses are getting intermingled in a non-deterministic way. Is there some way you can check for this condition in your setup?
The reason I say this is that you may think it is being handled ok by Tomcat but if this non-deterministic order is occurring and you have state going back and forth you may end up with correctness issues in your app. Maybe your app is not sensitive to this - if so if would be lucky.
[Message sent by forum member 'lwhite' (lwhite)]
http://forums.java.net/jive/thread.jspa?messageID=292861