From: Adrian B. (JIRA) <jir...@jb...> - 2005-03-14 21:05:59
|
[ http://jira.jboss.com/jira/browse/JBAS-1585?page=history ] Adrian Brock updated JBAS-1585: ------------------------------- Description: > -----Original Message----- > From: Adrian Brock > Sent: Monday, March 14, 2005 2:35 PM > To: Ivelin Ivanov > Cc: Stan Silvert; Neal Byrd; The Core; Rebecca Goldstein > Subject: RE: JMS Reference > ... > I know you can get about x8 improvement in JBossMQ just by adding > some simple flow control. > > Instead of doing the "one in, one out" policy on the message cache > when memory is tight (i.e. move some old messages to disk to make > room for the new messages) > I did a simple test instead with > Thread.sleep(1000); > to slow down the senders and let the receivers clear some memory. > > The message throughput went from 1 million messages/hour to 8 million. > I did this test around JBoss-3.0.7 time when the performance was > very poor. > > I discarded the idea as too simplistic since the mechanism > requires blocking server threads. But it does show JBossMQ > could easily be improved if we had the motivation/resources > to work on the performance without degrading its reliability. > was: > -----Original Message----- > From: Adrian Brock > Sent: Monday, March 14, 2005 2:35 PM > To: Ivelin Ivanov > Cc: Stan Silvert; Neal Byrd; The Core; Rebecca Goldstein > Subject: RE: JMS Reference > ... > I know you can get about x8 improvement in JBossMQ just by adding > some simple flow control. > > Instead of doing the "one in, one out" policy on the message cache > when memory is tight (i.e. move some old messages to disk to make > room for the new messages) > I did a simple test instead with > Thread.sleep(1000); > to slow down the senders and let the receivers clear some memory. > > The message throughput went from 1 million messages/hour to 8 million. > I did this test around JBoss-3.0.7 time when the performance was > very poor. > > I discarded the idea as too simplistic since the mechanism > requires blocking server threads. But it does show JBossMQ > could easily be improved if we had the motivation/resources > to work on the performance without degrading its reliability. > JBoss Forum Reference: http://jboss.org/index.html?module=bb&op=viewtopic&t=24658 Added link to forums thread where I said flow control should be integrated with the invoker level. > Enhance Queue throughput > ------------------------ > > Key: JBAS-1585 > URL: http://jira.jboss.com/jira/browse/JBAS-1585 > Project: JBoss Application Server > Type: Feature Request > Components: JMS service > Reporter: Ivelin Ivanov > > > > -----Original Message----- > > From: Adrian Brock > > Sent: Monday, March 14, 2005 2:35 PM > > To: Ivelin Ivanov > > Cc: Stan Silvert; Neal Byrd; The Core; Rebecca Goldstein > > Subject: RE: JMS Reference > > > ... > > I know you can get about x8 improvement in JBossMQ just by adding > > some simple flow control. > > > > Instead of doing the "one in, one out" policy on the message cache > > when memory is tight (i.e. move some old messages to disk to make > > room for the new messages) > > I did a simple test instead with > > Thread.sleep(1000); > > to slow down the senders and let the receivers clear some memory. > > > > The message throughput went from 1 million messages/hour to 8 million. > > I did this test around JBoss-3.0.7 time when the performance was > > very poor. > > > > I discarded the idea as too simplistic since the mechanism > > requires blocking server threads. But it does show JBossMQ > > could easily be improved if we had the motivation/resources > > to work on the performance without degrading its reliability. > > -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - If you want more information on JIRA, or have a bug to report see: http://www.atlassian.com/software/jira |