[rabbitmq-discuss] Perfomance Calculations?

[Michael Artz]

Is there any way to estimate the rough performance expectations for a given 
CPU/RAM combination?  We've been running some perf tests with RabbitMQ on 
AWS and have been getting some interesting results.  We started with micro 
instances (~400MB RAM, up to 1.2Ghz in burst), and saw sustained throughput 
rates of about 400 messages/second (16 byte messages), with the occasional 
spike up to 2K messages/second, which we attributed to AWS's CPU allocation 
algorithms.  On a personal machine (Mac Mini) we were seeing ~7K 
messages/second.  In both cases, the management API lists blocked_by as 
'flow'.

So, for capacity planning, we'd like to estimate the hardware (or VM size) 
that we would need to process a given number of messages/second via a topic 
exchange with at least one bound queue.  Do any numbers like this exist, or 
are we going to have to develop them ourselves?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20121010/61b23810/attachment.htm>