[rabbitmq-discuss] (no subject)

[Chris Chew]

Hi folks.

On Dec 7, 2010, at 8:07 AM, Marek Majkowski wrote:

> On Tue, Dec 7, 2010 at 13:55, David ODonnell
> <david.j.odonnell at hotmail.com> wrote:
>> The average latency for RabbitMQ seems very high compared to the minimum.
>> Are there any OS settings for Linux which could be set to improve those
>> values?
>> Latency would be critical for the application I am working on
> 
> Frankly speaking I have no idea. It might be JVM timing issues, it might
> be bug in MulticastMain, or network latency. Or maybe even RabbitMQ latency.
> 

More generally, I'm coming to realize that tuning the test client is one of the most important parts of performance testing and experimentation.

Along those lines I've found that using a combo of the "concurrent mark sweep" and the ergonomic "par new" collectors returned much more predictable results than using the VM defaults.

GC tuning seems to be a more like a black art than anything else.  But this article is what started me down the path:  http://java.sun.com/developer/technicalArticles/Programming/turbo/

Cheers,

Chris