Bug 777680 (SOA-192)

Date of First Response: 2007-12-01 05:11:58
project_key: SOA

This is mainly intended as a tracking issue!!!

1) I developed a small JMS performance test (https://svn.corp.jboss.com/repos/soa/trunk/qa/tests/framework/src/org/jboss/soa/esb/qa/framework/JMSPerformanceApplication.java). It was able to consume 100,000 messages in 172s. The helloworld quickstart is able to consume 50,000 (one zero less!) in 791s. This means that plain JMS is 9.2 times faster! And it was on full SOA-P server as well.

581 msgs per sec for plain JMS
63 msgs per sec for ESB (helloworld performance - no println aso.)

2) I filled queue/C with ESB AWARE messages (100,000, 5kB each). Then I started helloworld performance test with JMS listener (NOT gateway) connected to the queue/C and I let it process all the messages. It was able to process all the messages in 394s. Comparing to 50,000 in 791 with gateway in the play we have 4 times better performance (254 msg/s).

254 msgs per sec for ESB aware messages
63 msgs per sec with GW in the play

3) There's no major difference between JMS GW and JCA-based JMS GW - the bottleneck seems to be in the message serialization (see attached profiling report).

4) Processing the ESB aware messages is much faster than adding these messages to the queue. See attached profiling results - Util.serialize is the problem I think. The example was generating of 100,000 ESB aware messages to the queue/C. You can see the source code at https://svn.corp.jboss.com/repos/soa/trunk/qa/tests/framework/src/org/jboss/soa/esb/qa/framework/GenerateESBMessagesApplication.java

20,9% of time (114 s) was spent invoking this method javax.xml.parsers.DocumentBuilderFactory.newDocumentBuilder

What about creating a pool of DocumentBuilders (supposing they are reusable) or to use something different?