[rabbitmq-discuss] Performence issue with clustering setup

[Matthias Radestock]

Eugene,

On 24/11/11 10:29, Eugene Kirpichov wrote:
> Oh, that sucks. How large is the effect?

The answer is "it depends", of course :)

On 2.7.0, with hipe-compilation enabled, I can get a sustained rate of 
37kHz in a straight-through test (one producer, one queue, one 
consumer), with auto-ack, tiny messages, no persistence. [using our Java 
client's MulticastMain example with '-a -r 37000']. Turning on acks and 
persistence, I get a rate of 12kHz ['-f persistent -r 12000'].

If I enable confirms and throttle producers s.t. at most 1000 confirms 
are outstanding ['-a -c 1000'], I get 26kHz. And with acks and 
persistence ['-f persistent -c 1000'] this drops to about 3kHz.

Regards,

Matthias.