[rabbitmq-discuss] Fwd: Performance bottleneck on inter-node connection

[Matthias Radestock]

On 11/04/14 17:37, joseph rouphael wrote:
> Test 8: Dual node, Queues/Consumers on same node, 20 queue, ACK off
> -------------------------------------------------------------------
> Nodes: Double node
> Queues created on: Node1
> Producers on: node 2
> Consumers on: node 1
> Producer Processes: 20
> Producer regulate rate per process: 5KHZ
> Acknowledgment: OFF
> Expected rate: 100KHZ
> Achieved rate: Fluctuating between 10KHZ and 20KHZ
> CPU idle: 95%
> Note: Bottleneck on inter-node connection

What happens when you run the above test with both nodes on the same 
machine?

Also, you really shouldn't need 20 producers/consumers to saturate a 
broker, unless the machine has >20 cores. So I'd be interested to hear 
what the smallest producer/consumer count is at which you see a 
significant performance difference between the single node and dual node 
case.

Finally, it would be helpful if you ran your tests using RabbitMQ's 
standard performance testing tool, PerfTest - 
http://www.rabbitmq.com/java-tools.html#perftest. In order to replicate 
the above scenario you'd run 20 PerfTest instances with options
   -u q<n> -x 0 -a
and 20 instances with
   -u q<n> -y 0
with n in 1..20.


Regards,

Matthias.