Comments on: A Twitter Savior? http://broaddev.com/2008/07/03/a-twitter-savior/ Wed, 08 Sep 2010 10:51:41 +0000 http://wordpress.org/?v=2.5.1 By: alexis http://broaddev.com/2008/07/03/a-twitter-savior/#comment-256 alexis Fri, 04 Jul 2008 08:09:26 +0000 http://broaddev.com/?p=63#comment-256 Alex, thanks. We'd be happy to help - I think we have a good understanding of how to scale messaging systems. So if I understood correctly they were having problems with jabber and not Rabbit. By combining RabbitMQ with an XMPP gateway, we are able to provide scalable pub/sub (in twitter terms 'following friends posts') with an IM front end. This means twitter can offload work onto a set of RabbitMQ nodes. Potentially this could be done in a 'decentralised' way, though that would require federated roster. I'd be happy to chat more if you contact me. Cheers, alexis Alex, thanks.
We’d be happy to help - I think we have a good understanding of how to scale messaging systems.
So if I understood correctly they were having problems with jabber and not Rabbit. By combining RabbitMQ with an XMPP gateway, we are able to provide scalable pub/sub (in twitter terms ‘following friends posts’) with an IM front end. This means twitter can offload work onto a set of RabbitMQ nodes. Potentially this could be done in a ‘decentralised’ way, though that would require federated roster. I’d be happy to chat more if you contact me. Cheers, alexis

]]> By: Alex Lewis http://broaddev.com/2008/07/03/a-twitter-savior/#comment-253 Alex Lewis Thu, 03 Jul 2008 21:12:02 +0000 http://broaddev.com/?p=63#comment-253 Hi Alexis, I have a feeling it was too few clusters/systems for an environment of ~15,000-20,000 concurrent users. It was a tangent to a different project I was working on. I can't mention their name but I know they currently only use it in a limited fashion and have implemented a more standard corporate IM solution instead of the somewhat custom jabber/Rabbit solution. Do you have any thoughts on how the Rabbit project might affect/assist Twitter? I know they could use the help and are looking. Hi Alexis,

I have a feeling it was too few clusters/systems for an environment of ~15,000-20,000 concurrent users. It was a tangent to a different project I was working on. I can’t mention their name but I know they currently only use it in a limited fashion and have implemented a more standard corporate IM solution instead of the somewhat custom jabber/Rabbit solution.

Do you have any thoughts on how the Rabbit project might affect/assist Twitter? I know they could use the help and are looking.

]]> By: alexis http://broaddev.com/2008/07/03/a-twitter-savior/#comment-252 alexis Thu, 03 Jul 2008 20:56:53 +0000 http://broaddev.com/?p=63#comment-252 Alex, what was the problem? Please let us know if it was Rabbit related. In our tests a RabbitMQ node can handle about 4,000 concurrent TCP connnections, and by clustering RabbitMQ you can handle many more. If you connect via a connectionless protocol such as HTTP, RabbitMQ can handle much larger numbers of concurrent users. In tests each node can handle hundreds of thousands of queues. If you would like help achieving these numbers stably, please tell me. alexis Alex, what was the problem? Please let us know if it was Rabbit related. In our tests a RabbitMQ node can handle about 4,000 concurrent TCP connnections, and by clustering RabbitMQ you can handle many more. If you connect via a connectionless protocol such as HTTP, RabbitMQ can handle much larger numbers of concurrent users. In tests each node can handle hundreds of thousands of queues. If you would like help achieving these numbers stably, please tell me.

alexis

]]>