On 5/30/06, jordan woerndle <[EMAIL PROTECTED]> wrote:
I'd like to start out by pleading JMS ignorance and apologizing for my generic question... I'm really just looking for some suggestions on how I can get better performance out of a wacky activeMQ app. I have a bunch of "Filters" that listen for and broadcast messages on many Topics. I want each filter to be able to listen to many topics, and periodically fire a message off onto one-to-many topics. It kind of looks like a network of neurons passing messages around and making decisions (think distributed bayesian filter.) Currently, each of my filters subscribes to one topic. Each will periodically publish a message to one other topic and send a "status" messsage to a "status" topic. I can run about 30 "Filters" (they are doing no processing) with about 30 topics.. i can push about 90 messages a second. I'd like to do a lot more and listen,broadcast to more topics.
So is that 90 mess/sec * 30 filters = 2700 msg/sec total? With topics we typically see performance in the order of 1000's of messages/sec. Are you CPU constrained or IO?
I am running activeMQ 4.0 M4 using non-persistent messaging, windows XP, jdk1.5.0 on one machine using one broker started from the command line using the default out-of-the-box configuraiton. Everything I know i learned from the examples on the site. If it helps, I don't really care if I drop a few messages.
I would upgrade to the listed at latest http://incubator.apache.org/activemq/download.html
What's the first step to increase throughput? Should I put the broker on a different machine? Should I try to split my topics up between several brokers?
Splitting the topics sometimes helps.. You can also decrease CPU usage by avoiding marshalling by using the vm:// transport.
Thanks a lot! Jordan -- View this message in context: http://www.nabble.com/neural+net+using+topics--t1707208.html#a4634774 Sent from the ActiveMQ - User forum at Nabble.com.
-- Regards, Hiram
