We have a web application publishing persistent messages to a virtual topic within XA transactions. Message listeners remove messages from their queues almost as soon as they arrive. We're running ActiveMQ 5.4.2 using KahaDB persistence.
The system performs well under varying simulated user loads running against a single ActiveMQ broker. Running against a no share master/slave broker configuration, transaction times increase by 2 seconds after doubling the number of users. We've been unsuccessful at profiling the master and slave brokers using Wylie to find where the extra time is being spent. There are so many threads active for so many classes that I don't know where to start. Any suggestions for how to approach this problem? -- View this message in context: http://activemq.2283324.n4.nabble.com/xa-transaction-times-double-with-no-share-master-slave-ActiveMQ-tp3731476p3731476.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
