[Rhm-users] RHM performance

[Andy Grove]

Hi

 

I've just been running some benchmarks against the RHM broker and the Apache
QPID Java broker. They are showing very different performance and
scalability characteristics and I'm rather surprised to see that the RHM C++
broker is many times slower than the Apache QPID Java broker when running
these tests.  I'm hoping that this is just a configuration issue. I've
posted a graph of the results in PDF format at the following location:

 

http://www.codesuccess.com/temp/rhm_qpid_perf.pdf

 

Can anyone shed any light on these results? I was assuming that performance
would be better with the RHM C++ broker.

 

Here's a brief description of the tests:

 

The test consists of client and a server. 

 

The server creates a durable subscriber for 50 topics and listens for
messages (non transactional, auto acknowledge).

 

The client that starts with a single thread and then adds a new thread every
5 seconds. Each thread continuously publishes a fixed size message to one of
50 topics at random. Throughput is measured on the client for each 5 second
duration (the cost of setting up each thread and connecting to JMS is
excluded from the throughput measurement). The client uses persistent
messaging and transactions.

 

The test client, test server and broker are each running on a separate quad
core 2.4GHz Intel PC with 4GBM RAM running Fedora 8 (64 bit).  Each machine
has an identical hardware specification. The PCs are all connected to the
same gigabit switch.

 

I used the same client java jars (from the RHM RPM) when testing against the
RHM C++ broker and the Apache QPID Java broker.

 

The RHM broker was not configured with a persistent store but QPID was. RHM
and QPID brokers were run on the same server, but not at the same time.

 

Thanks,

 

Andy.

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/rhm-users/attachments/20071208/e45b52f9/attachment.htm>