Oleg
On Jun 1, 2010, at 7:09 PM, Matthew Sackman wrote:

> Hi Bradford,
> 
> On Tue, Jun 01, 2010 at 03:27:51PM -0700, Bradford Stephens wrote:
>> I'm having pretty nasty speed issues on Fedora. I'm running Fedora 8,
>> 64-bit on EC2. I have a 8core/8GB RAM server. I'm able to insert
>> thousands of messages a second, but I can only consume about 3/sec.
>> Each is ~1k.
> 
> Well for starters, don't use basic.get, set up a consumer using
> basic.consume. Secondly, think about whether you need to acknowledge
> messages - if they're relatively unimportant messages (i.e. if a client
> crashes whilst processing messages, do you want those messages to be
> lost, or to automatically be requeued) then you might consider turning
> on noAck (looks like you already do in your example code), which helps
> speed.
> 
> Finally, you're bascially writing a poor version of the QueueingConsumer
> (the major difference being that QueueingConsumer permits a jump across
> threads, meaning that you can process received messages in a different
> thread, and not impact the rate at which you receive messages from the
> broker). I'd recommend you try using the QueueingConsumer instead as it
> would then mean that you don't block the client whilst you're processing
> the messages (in this case just adding messages into a queue -- err
> which is what the QueueingConsumer does for you automatically).
> 
> As always with performance issues, once you're using the right API and
> right algorithms (and asking here is definitely the right thing to do),
> careful use of a profiler is a good idea to pick up where any further
> bottlenecks are.
> 
> Matthew
> 
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-discuss@lists.rabbitmq.com
> http://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss