I am writing a remote backup service meant to support 1000+ users. It is going to use librsync to store reverse diffs (like rdiff-backup) and make data transfer efficient.

My trouble is that I do not know the "best" way to implement the client/server model. I have thought of doing it like rsync/rdiff-backup do it by having the client open an SSH connection and running a server executable and communicating across pipes.

Another alternative would be to write a server which would handle authentication and communicate with the client via SSL. The reason I have thought of this is that there is "state" information like how many backup jobs are setup, etc. that must be maintained.

Another alternative that I have thought about is running a "web service" using Pylons or Django to handle the authentication, but I do not know how to bridge that the the "storage" side. Since I am using librsync, I cannot use "dumb" storage. Is there a way to pipe data through Pylons or Django to a server side handler that would do the rsync calculation?

This seems to me like maybe a dumb question but I am sort of lost. Any tips or suggestions from more experienced developers would be extremely helpful.