[rabbitmq-discuss] Messaging at SecondLife

Ryan Williams (Which) rdw at lindenlab.com
Tue Apr 7 22:50:20 BST 2009

Alexis Richardson wrote:
> 2009/4/6 Martin Sustrik <sustrik at imatix.com>:
>> Holger Hoffstätte wrote:
>>> SecondLife tried to evaluate several messaging solutions and a writeup of
>>> the (quite sobering) results can be found here:
>>> http://wiki.secondlife.com/wiki/Message_Queue_Evaluation_Notes
>> Their scenario (load-balancing queues across several network locations)
>> is pretty exotic. It's no surprise that none of the messaging systems
>> supports it out of the box. AFAIK even MQ series or Tibco doesn't
>> support this kind of thing. Interesting article though.
> Indeed.
> They actually have several use cases, not all of which are documented
> here.  It's a very well structured analysis though.
> If you extra nice to Brian Sullivan he may chat about some of their
> use cases on this list.

I'm the horses's mouth for that wiki page (and the horse's ass too!).
We focused on the group chat use case pretty much to the exclusion of
all others because it is the hardest to solve and would provide the most
differentiation amongst implementations.  And it really served that
purpose well, seeing as how only 3 implementations even approached a
passable solution.

I started a thread about our intended usage here:
Which led me to produce the section on scaling.  We were definitely
hoping to stumble over some software that just did what we wanted, but,
since nothing appears to do so, that's an interesting negative result in
and of itself.  It was not such an obvious conclusion for us.

Now we're left with a bunch of options:
 - Improve Rabbit or some other message queue software so it's closer to
what we want.
 - Change some fundamental constants of the chat system so it is no
longer quite so esoteric.
 - Relegate group chat and IM to special-purpose software and use
message brokers for everything else.
 - Keep on looking for software that solves our use cases out of the box.

I'm not sure what the correct course of action is.  I believe that it is
technically possible to build the system we want, so I hold out hope
that it will be solved somewhere, somewhen.

More information about the rabbitmq-discuss mailing list