JBossDeveloper

Hi Jaikiran,

thank you for fast response again. I adjusted the option, but the effect was minimal. In one or two runs I had a performance gain of ~1MB/s when I set the option on the client and on the server side to 10MB. But again, this doesn't reach the bandwidth of our connection. The profiler shows the same picture as before, a lot of wait time within org.jboss.remoting3.remote.OutboundMessage$1.accept(org.xnio.Pooled, boolean).

Regards

Martin

Hi Jaikiran,

in hope to get a solution, I have created a JIRA issue for this problem: https://issues.jboss.org/browse/EJBCLIENT-75.

Regards

Martin

Hi Jaikiran,

thank you for returning to this thread.

I have commited my standalone.xml to the github repo: https://github.com/mmois/jboss-ejb-client-performance/blob/cde8252466d48bca358dba06c1996dc9ab94be5d/server/jboss.config/standalone.xml.

The following snapshot shows the stack trace of one of my threads, with huge differences between CPU time and execution time:

The next snapshot displays the "hot spots":

Hello Jaikiran,

have you made some measurements by varying the number of threads or the message size? Here are some examples from my computer with your updated configuration:

- 100 iterations,  1 Thread , 1MB message size: 18,573551 MB/s.

- 100 iterations,  5 Threads, 1MB message size: 28,130978 MB/s

- 100 iterations, 10 Threads, 1MB message size: 28,797696 MB/s

- 100 iterations, 15 Threads, 1MB message size: 29,388715 MB/s

- 100 iterations, 20 Threads, 1MB message size: 29,129890 MB/s

- 100 iterations, 25 Threads, 1MB message size: 28,325723 MB/s

- 100 iterations,  1 Thread, 3MB message size: 23,557126 MB/s

- 100 iterations,  5 Threads, 3MB message size: 29,619098 MB/s

- 100 iterations, 10 Threads, 3MB message size: 28,561908 MB/s

- 100 iterations, 15 Threads, 3MB message size: 29,250219 MB/s

- 100 iterations, 20 Threads, 3MB message size: 28,974170 MB/s

As the figures show, there seems to be a specific limit, which you cannot overcome by adding further threads or by increasing the message size. But when I use iperf (http://sourceforge.net/projects/iperf/) to measure the bandwidth of my loopback interface, I see that 72,3 MB/s are possible. Thus, the measured bandwidth is about the factor 2 away from the practical limit measured with iperf. This is why i stated that there seems to be some kind of saturation.

Can you measure the bandwidth of your network connection and perform some tests by varying the number of threads and the message size and post your results here?

Regards

Martin

Just to underline my last measurements, here a screenshot of Window's Task Manager showing the network throughput on our gigabit connection while running the tool with even higher values than you used:

Regards

Martin

Hi Jukka,

we haven't reached any conclusion yet. My JIRA issue has been closed and the response to Jaikiran's last posting is still outstanding.

In the meantime we have also investigated a clustered setup. In this case the jboss-ejb-client library will dispatch the remote calls to both application servers, but the limitation seen with one application server still remains. Thus, the problem must definitely lie on the client side.

Maybe you can reopen the JIRA issue and describe in a comment your problem?

Regards

Martin