First of all, for your full-round-trip case (no production of message n+1 until triggered by result returned for message n), you need to know the network latency. So do traceroute or equivalent both ways. US->AUS and AUS->US.
This gives you a hard limit on how many round-trips you can do per unit time - i.e. your maximum possible performance for that case. I suggest that it will be *way below* 100 roundtrips/sec. You could also turn on full logging: For each of your brokers, edit the log4j.properties file for debugging (as indicated in the file itself), and push the button in e.g. jconsole to reread the log4j.properties file. Mbeans tab -> org.apache.activemq -> broker_instance_name -> Broker -> Operations -> *reloadLog4jProperties * Finally, you could remove one broker at each site to help understand this. For sure, they're not helping your performance. ----- Michael Hayes B.Sc. (NUI), M.Sc. (DCU), SCSA SCNA -- View this message in context: http://activemq.2283324.n4.nabble.com/Consumer-performance-improves-after-bounce-when-in-cluster-setup-tp4514510p4514812.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.