Clearly I have been having some serious misunderstandings on how this works. 1. My clients use the JCA adapter because we develop on windows machines that can run a windows only conversion library, so the "ConversionJMSProvider" is defined to connect to localhost on our machines and to connect to a cluster on the deployment environment. Our deployment environment has only the conversion service running on a windows JBoss, which connects to a cluster of linux JBoss AS. I thought this was the way to do that. 2. anonymous wrote : java.naming.provider.url=jnp://devapp2.qa.cin.int:1100,jnp://devapp1.qa.cin.int:1100 I was under the (mis) understanding that this was how you specified the servers in the cluster. We were initially having issues with discovery. I think that it was related to firewall settings or something. 3. I (mistakenly, apparently) thought that a producer that uses the ClusteredConnectionFactory would round robin the messages as they were posted. We have a service that runs as a singleton that produces messages and we want them to be round-robined to the consumers running on each node. It seemed reasonable to me. After running a series of tests I see that virtually all of my assumptions were ass-backward. Time to rethink my strategy. View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4144479#... Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&a...