Performance not satisfactory with IBM JMS Perf Harness

[sureshachary]

I've been trying to measure the performance of various JMS providers using IBM JMS Perf Harness testing tool (available at http://www.alphaworks.ibm.com/tech/perfharness). But the numbers (22 msgs/sec) I got was not fair enough to choose HornetQ.

Did anyone try measuring the performance of HornetQ using IBM JMS Perf Harness testing tool and got any fair number?

I went with the default settings of HornetQ and didnt do any customization.

So can anyone help me with any configuration to get a fair performance with HornetQ and do  have to use any special settings when I use IBM JMS Perf Harness.

Thank you.

Suresh B

[ataylor]

first of all if you are comparing HornetQ against other JMS vendors then its like comparing apples and oranges, some vendors such as JBM dont use proper persistence out of the box and other vendors won't sync persistant messages to disk by default, take a look at the HornetQ technical FAQ theres some info there, http://community.jboss.org/wiki/HornetQTechnicalFAQ .

Also what would you call a fair number, when using strict out of the box synching and acknowledging then this is typically constrained by the network bandwidth and disk.

Theres also a section in the user manual on performance tuning that may be of help to you.

[clebert.suconic]

+1 on Andy:

Most likely you are syncing on HornetQ and not syncing on the other system you are choosing.

You should compare oranges to oranges...  i.e.  syncs with syncs.

We can take a look if you prove there is an issue after you do the proper and valid tests.

[sureshachary]

I've tested with async & persistent messages while testing all the messaging servers.

I compared HornetQ, WebsphereMQ, Weblogic JMS, Websphere App Server JMS, Qpid etc with the same settings.

But with Hornet, I got only 22 msgs/sec and with others it was in hundreds.

Thats why I requested whether there is any special settings to be done.

The client and the server were running on the same machine in all the testing scenarios.

If you could please dowload the IBM JMS Perf Harness utility(download link provided above) and test you can see the difference.

[ataylor]

when we talk about syncing we are talking about syncing to disk when messages are persisted, not all, in fact most, messaging vendors don't do this and therefore don't out of the box support full recovery. If you want to relax thses semantics on HornetQ the user manual explains how to do it.

Sorry, im afraid we dont have time to do your perf testing for you, but if you find any issues let us know.

[sureshachary]

fine, but it just takes 15 mins to setup and start the testing.

and the issues i've already posted.

[clebert.suconic]

I did the comparisons to these servers myself.

All of them are optimized for benchmarks. i.e they don't sync.. and not for production systems.

Make sure they are syncing to disks... and making proper roundtrips when committing and sending persistent messages.

>>fine, but it just takes 15 mins to setup and start the testing. and the issues i've already posted.

You didn't post any issues.. you are just comparing a system optimized to not lose messages (hornetq) against a system that's optimized for pointless benchmarks.

You should do the proper settings.