>=20 > Andy! >=20 > frankly - why would I create value objects in the session bean? Of course= they > are created in ejbLoad. >=20
Good. This substantially effects performance on eager loaded and option A caches.=20 What is your container config? Can you post your descriptor (ejb-jar.xml, jboss.xml and jbosscmp-jdbc.xml)? I want to see what fields you're eager loading as well as what you're caching, marking read only (to prevent transaction locking) and what locking policies you're using (if collisions are rare then do you need pessimistic locking?) >> Lastly, Was that WAS on IBM JDK and JBoss on Sun JDK? Sun's JDK >> is probably the slowest and has a nasty classloading bug that you're ver= y >> likely to hit on JBoss 3.2.2 and below under load and might hit on 3.2.3= . >> (It is never fair to compare on two different VMs.) >=20 > You are right, JBoss was runnning in a Sun JDK, WAS on the IBM JDK (both > RedHat AS 3.0). And the main problem we had was excessive garbage collect= ion > that eventually lead to a crash (which made me advocate using a different= JDK > like JRockit or the IBM JDK - the tests just haven't happened yet). > Right, that=B9s totally not a fair comparison. How does JBoss do on the IBM JDK? Which JDK version? Do you have multiple processors? Can you use the low-pause garbage collection? The default garbage collector is single threaded and stops all threads (essentially) and thus performs suboptimal on multi-processor systems. That you have excessive garbage collection especially makes me wonder how you're making use of transactional and second-level cache. (We'll talk about all of this in great detail in the august training and less in the august jug talk) =20 > I didn't mean to badmouth either application server (at least not in this > particular mail :). Both have their purpose and application. Basically I = was > just pleased to say that after all the hassle I had to go through, gettin= g > things to run on WebSphere, there were actually reasonable results. >=20 > However - you got me interested in that classloading bug you mentioned. W= ould > you care to elaborate or give me a pointer so that I can read it up? > http://developer.java.sun.com/developer/bugParade/bugs/4699981.html IBM shouldn't ask Sun to open source their JDK, IBM should open source theirs. ;-) Supposedly Jrockit is very fast but my test server is still running an old version of redhat and it was DOOOOOOOGGGGGG slow for me. I think it was my version of Linux's threading though. Generally most Jboss guys recommend Jrockit because it is just that good. I'm planning to upgrade as soon as I pick a new linux distro (red hat pissed me off, I don=B9t want to use it on principle -- almost makes me want to become a socialist!). -Andy =20 > Thanks, >=20 > -hendrik >=20 > _______________________________________________ > Juglist mailing list > [EMAIL PROTECTED] > http://trijug.org/mailman/listinfo/juglist_trijug.org >=20 _______________________________________________ Juglist mailing list [EMAIL PROTECTED] http://trijug.org/mailman/listinfo/juglist_trijug.org
