I hava done some test while using the new feature Replicated LevelDB. I'm running a 3 node ActiveMQ quorum_mem and 3 node Zookeeper. In a PtoP situation,using a pooled connectionFactory to send Messages with one queue. I observed 3000 msgs/sec which is similar with using levelDB or KahaDb in normal situation. However,the performance goes down to 100 msgs/sec while the queue is empty and has consumers waiting for receiving.
It's really strange.This only occurs when the queue has consumers waiting for receiving and the queue is empty. And in my test the only one way to improve the throughput is to increase the maxConnections in the pool. It seems that the throughput of a connection is restricted while it's all ok if i changed the persistAdapter to levelDB or kahaDb. I doubt there is lock problems when dealing with sending and receiving at the same time for replicated levelDB. I'm very confused now.I wonder whether it's my configuration problem or it's really a bug. And i wonder which master-slave solution is maturer and recommended for activemq in a industrial practice. -- View this message in context: http://activemq.2283324.n4.nabble.com/Replicated-LevelDB-performance-problem-tp4679150.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.