Hi, We are testing 5.5.0 for an upgrade from 5.4 - hoping for a performance boost.
In our testing, we notice brokers stop delivering messages on a regular interval. The cpu graph clearly showing a slowing down of the broker at regular interval. With some investigation by changing the kahaDb's cleanupInterval and DEBUG trace, we think we have identify where the "pause" occurs. For simplicity, we reverted our activemq.xml to minimal config. Our test client is sending at approximately 39k msgs/sec to a test.queue. With debug trace, the pause is between timestamp 2011-07-29 22:07:38,617 through 2011-07-29 22:08:03,214. During that period, no message were delivered for test.queue and broker's cpu usage drops to 0 (jconsole). The trace is in the attachment. It looked like the pauses corresponding to kahaDB's cleanupInterval. The trace showed the following activities during the pause interval: - Main:memory: usage change from: 36% of available memory .... at interval of 1% points - org.apache.activemq.transport.InactivityMonitor$2@7aaea288 10000 ms elapsed since last write check - Multiple Checkpoint started.done. - expiring queue messages. We saw the same queued were checked multiple times w/in the interval. Any help to eliminate the pauses is greatly appreciated. Thanks
