A friend and I started working on a prototype distributed file system to run on our lab machines. We have about 40 machines, most of which are PowerMac G5s, with a total of 18TB of available disk space. The idea is to create an abstract high-level filesystem on top of which we can build proxies that implement various data retrieval protocols like FTP, HTTP, and AFP. We're writing the prototype in Python, but we're also using things like Rendezvous (zeroconf) for node auto-detection, which is quite exciting. We still have quite a bit to figure out, but there are numerous fun problems to solve, and the prospects of being able to give each student hundreds of megabytes of storage (with everything stored in triplicate) makes the project also very worthwhile... if we can do it, of course.

This project spawned from our discussion about installing IlohaMail on our lab machines, and how cool it would be if we could parallelize it to the point where we can just include an instance of IlohaMail in the lab images, and have the entire lab turn into one giant application. With all the dual-G5s, we could easily support tens of thousands of users, without having locally logged in users notice any performance degredation.