Are you saying that running stress testing under JMeter does not cause it to go to noddy land? i.e. the behavior is not reproduced in 'controlled conditions'?
Yup. I've used JMeter and JProfiler successfully in weeding out some issues with synchronization, but those should be gone with 2.6.
In that case I would be looking at what's the last thing in the logs before it freezes. Any patterns there?
Not that I can see. It is probably a combination of timings, and also some effects from other load of the computer (e.g. lots of spam mail coming in => spamassassin runs => high amount of disk access => lots of load)
Otherwise, even using the Apache logs might not be enough because (presumably) that would not contain the actual content of form posts.
Yes.
Sounds like a big ask on your box!! Does not Apache Incubator offer hosting you can take advantage of?
I would certainly hope so. /Janne
