Sounds strange. From my expercience JBossMQ is faster that that, but probably not 500 msg/sec. I would say 10ms per message would not be to uncommon wich would make 100ms/s the normal throughput.

//Peter

I assume you're letting the test run long enough
to be sure you're not measuring overhead from
class-loading, JIT invocation, etc.

The simple testing I've done showed response
time in the 10-30 ms range. Perhaps there is
some particular aspect of your test which is
exposing some inefficiency in JBossMQ. For
example, are you using message selectors?
Does receive() make a difference compared
to receive(timeout)? Are the messages
themselves large?

I am also facing the performace issue. My requirment is minimal and it is 10 messages per second. I am using non durable topics with MDBs and web side clients as consumers. Message flow is basically a stock quote feed from a server.

In my tests the Jboss consumes all of the memory with in few hours (I have 512 MB RAM) and then start the caching the message on disk.

Am I doing any thing worng in configuring the JBoss? Why the messages are not purged after consumption by the subsribers?

Any help will be apriciated.

Raj

Were you using selectors??? A recent bug has been found that might have fixed your problem.

Regards,
Hiram

Have you tried to use IBM VM for Linux. I heard it is _much_ fatser than Sun's?

I have the same issue. The performance of JBossMQ on Windows and on a Sun Solaris machine are significantly
different. Here is the benchmark I did:

1 publisher, persistent mode, publish 512-byte text message as soon as possible to a topic. The publisher program runs
on a separate JVM from the JBoss/JBossMQ server.
The default RollingLogger is used.

5 durable subscribers, each using a unique connection and
session. All 5 subscribers running in 5 different threads of a JVM that is different from the JBoss/JBossMQ server.
After recieved an message, it drops it as soon as possible
so that there is little processing there.

The throughput is measured solely on the subscriber side
over a time of 10 minutes.
JBoss is 2.4.4, SUN JDK 1.3.1_02.

On an Intel Pentimum III, single CPU, 700 MHz, 512MB RAM, Windows 2000, the throughput is ~400 messages/s. I observed ~90% CPU ultilization during the run.

On a Sun E220R, dual CPUs, 1GB RAM, the throughput is
~50 messages/s. I saw only 4% CPU ultilization.

Can someone explain why the performance/throughput is
so different ? I have done a similar benchmarking
on the JBoss EJB components on the two systems, the difference is very little.

Thanks.

--- Gang

we faced a similar problem, the performance improved when we used the tcp_nodelay option (Niguels algorithm0added in jboss 3.0 .

we faced a similar problem, the performance improved when we used the tcp_nodelay option (Niguels algorithm0added in jboss 3.0 .

How to use the tcp_nodelay option ? Can you configure it
via Solaris or JBoss 3 settings ? Thanks.

--- Gang

Can anyone explain how to using JBossMQ into jBoss 2.4.4. How is the config etc done