[rabbitmq-discuss] Active Queue Management

[Matthew Sackman]

On Thu, May 10, 2012 at 08:51:24AM -0400, Tony Garnock-Jones wrote:
> It's interesting to note that the same queue-management issues that appear
> on the Internet at large crop up not only within brokers like RabbitMQ
> themselves, but also within messaging client libraries. (You are trapped in
> a maze of congested queues, all different.) Ad-hoc solutions scale quite a
> long way inside a process, and not nearly so far across process boundaries;
> I wonder if implementing some "real" AQM within a process, e.g. in client
> libraries or within brokers, would help at all? Maybe for clustering and
> federation.

I'm actually in the process of implementing CoDel in the Java client.
What it should allow for (I hope - unless I've misunderstood something)
is to be able to not set a QoS or, set a pretty arbitrary QoS, but then
the algorithm ensures the client-side buffer doesn't just remain full.
The assumption is that the transition from "delivery to the client" to
"processing the message on the client" is the fast->slow bandwidth
transition that AQM is meant to manage.

But I may be wrong on one or more of these assumptions!

Matthew