I have been running some tests to test the performance of JBoss 3.0 versus that of 
JBoss 2.4 and have found that, under load, JBoss 3.0 is considerably slower than JBoss 
2.4.  My initial posts on this topic were to the DB forum but Dain Sundstrom suggested 
that I post my findings here.  The original posts are at

http://www.jboss.org/forums/thread.jsp?forum=46&thread=13695

Since my original post I found one other problem which is not a big contributor but is 
an easy fix.  For some reason, in org.jboss.tm.TxManager, the trace member is set to 
log.isDebugEnabled() which returns true.  This causes TxManager to attempt to log 
requests.  This is problematic because (a) the log messages don't end up anywhere (at 
least they aren't in server.log) and (b) unnecessary calls to 
org.jboss.tm.XidImpl.toString() are made which contributes a small but not 
insignificant amount to the performance problem.  By comparing to 
org.jboss.tm.TxCapsule, I think that trace should be set to log.isTraceEnabled().  

By studying the OptimizeIT snapshots a little more, I've come to the conclusion that 
while the unmarshalling process seems to be taking more time than it should, it is 
probably not the whole story and there are other sources of the performance hit as 
well.  Unfortunately, I cannot see anything else so obvious in the OptimizeIT output 
other than everything seems to take longer especially as the server load goes up.  
Hopefully someone with more familiarity with the code than I have and that has an 
OptimizeIT licence that is not about to expire can take a closer look at this issue.

Dan.

* * *

View thread online: http://jboss.org/forums/thread.jsp?forum=66&thread=14367

_______________________________________________
Jboss-development mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/jboss-development

Reply via email to