[rabbitmq-discuss] couple of questions (was: Re: RabbitMQ 2.6.0 is released)

[Marek Majkowski]

Murthy,

On Fri, Sep 2, 2011 at 00:54, Murthy Jayanthi <mjayanthi at gmail.com> wrote:
> I am investing light weight. stable messaging products and RabbitMQ looks
> great to me. I have a couple of questions and I am hoping that someone could
> help me get answers.
>
> 1. Does RabbitMQ support  an external persistence adapter for persist
> message delivery to avoid message guaranty.
> 2. The idea is to use Cassandra replicated storage for managing the persist
> messages.

Currently, RabbitMQ uses its own, well tuned, persistence layer.
It's possible to replace current implementation with a different
one, few attempts were made to support alternative  data stores,
but nothing reached GA.

RabbitMQ is generic message broker which supports many
messaging patterns, it would be unlikely that a simple k-v message
store (like cassandra) would behave performant with load patterns
that RabbitMQ creates.

> 3. If anyone has tried that, what is the ballpark throughput of a single
> instance of Rabbit?

This is an open-ended question.
 - Do you care about reliability (persistence)?
 - Do you need transactions?
 - In general, are your consumers going to be faster
  than producers? (ie: are you going to queue messages at all?)
 - What size of the messages can you expect?
 - What routing topology?
 - What number of queues is going to be used, and what
   will be their load?

If you want a single number - our tool - multicastmain,
documented for example at the bottom of this page:
  http://www.rabbitmq.com/ec2.html
Shows up to 30k msgs/sec in a simple scenario
(but in practice this is a meaningless number).

What we usually suggest, is to write a prototype of
your application and do end-to-end tests.
I would be surprised if RabbitMQ appeared to be
the bottleneck in your app.

Cheers,
  Marek