Overview
This is a presentation that explains why OpenAMQ does not do persistent messaging. The short story is that while other AMQP brokers do reliability using transactions, persistence, and so on, we don't think this gives us the performance and reliability our users need. Why settle for 3,000 messages a second when you could get 150,000? Classic broker designs do reliability from client to broker, and broker to client. We do it from client to client, so-called end-to-end reliability. The broker becomes an asynchronous message switch or AMS.
To see why AMS offers much better scalability, while being much simpler than traditional centralised reliability, read the presentation.
State of play
The techniques described in this presentation have been tested and used in very large applications. We know they work. They work on AMQP/0.8 and need no reliability support in the protocol itself. The actual implementation is in the client API framework, not the protocol and broker. The iMatix AMS stack is not currently release-ready; the designs documented here should be treated as a design proposal, not documentation of a fully-implemented system.
License
The presentation is licensed under the Creative Commons Attribution-Share Alike License 2.5.
Support
- For all questions, contact the author Pieter Hintjens on moc.xitami|hp#moc.xitami|hp.