We are using Orion merely as a servlet server  (no ejbs) and we are hosting
surveys (forms, questionaires) in it.  We have built the validation /
routing logic into the server using rhino (js.jar) and we have our own
proprietary data store.  

When we go to test the server under stress (as much load as we can get it to
take simulating 30-50 users hitting it as hard as they can) after about 14
hours using hotspot (Sun's server VM) and 112 hours using classic it either
crashes or locks up.

The hotspot VM dies with and internal error and the VM just quits.  I have
seen this VM allocate over 3000 threads.

The classic VM runs Windows NT out of virtual memory. (lots of threads here
to)

Has anyone seen anything like this ?

We are a software development company and are using many different
configurations so far the servers we ahve tested are Windows NT4 (with SP 5
or SP 6) and W2K (both server versions) and we are using orion 1.3.8 and
js.jar (rhino 1.5).  

I don't think we can come up with a "little test" to send to Sun as bug
against their VM - at least Orion is needed and that's not little.  

Does anyone have any ideas about how to make our server stay up under
stress.

Nate

Reply via email to