My advice is to simplify your environment, application, and broker configuration as much as possible while still reproducing the high CPU utilization. If you can do it with only local resources (e.g. everything running on a laptop) then that's even better. The simpler and easier it is to reproduce the faster (and more accurately) the problem can be analyzed and resolved.
After simplifying, the first thing would be to determine if the high CPU is being caused by garbage collection or something else. You can monitor heap utilization with tools like JConsole and VisualVM. Timely thread dumps are a great way to see what's consuming CPU within the broker itself. Good luck. Justin ----- Original Message ----- From: "francesco81" <francesco.padov...@bticino.it> To: users@activemq.apache.org Sent: Thursday, April 6, 2017 10:08:03 AM Subject: Re: Apache Artemis - Stress-test time Hi Justin, we've tried by turning off tls. It doesn't change: cpu usage remain very high. We've tried also by disabling persistence. It changes a bit: at the beginning of the connection phase, cpu goes to 50% ...then it increases time by time til it reaches 100% when there are about 1000 connected clients. Now we're going to add some logs to our scecurity plugin, to see the cpu consumption for every involved thread. Meanwhile I'm going to install and use Sun Oracle JDK 8, instead of the current OpenJDK 8, but simply to have jmap, jstack and jcmd utilities to make some additional analysis (I can't find them inside the standard OpenJDK package for Centos7). I'll let you know for further results. Of course, any suggestion to make a better performance analysis is appreciated. Thanks a lot. Francesco -- View this message in context: http://activemq.2283324.n4.nabble.com/Apache-Artemis-Stress-test-time-tp4723999p4724632.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
