It is federated in a manner similar to traditional email. There are many different servers with many users on each one, it's straightforward to run your own se

It is federated in a manner similar to traditional email. There are many different servers with many users on each one, it's straightforward to run your own server, and users on server A can easily communicate with users on server B. If your server goes down for some reason, much like email, you'll lose the ability to log in and post, but anyone on a different server will be unaffected. I don't believe there's any particularly special DB replication features in the reference server implementation, but there's nothing inherent in the system that would prevent someone writing a customized durable/distributed version of the server.