AFS solved the cross OS permission problem by throwing away OS permissions and handling permissions itself entirely. It's tightly integrated with kerberos for auth/encryption enabling trivial federation.<p>It also solved a lot of data locality/ discovery issues by using SRV records. You want to visit /afs/cmu.edu? No problem, client looks up the dns record then uses that to ask the volserver where the volume actually lives.<p>The openafs server implementation is single threaded unfortunately. However is a more performant proprietary implementation via auristor (it's not all bad as Jeff Altman has been involved in openafs for a long time).<p>If you don't want to install the openafs kernel module on all your clients, there is also a stripped down in-kernel Linux client, kafs.<p>You still run into it in the research/edu space although a lot of places now block outbound afs so you don't accidentally leak data.<p>I highly recommend checking out other software that came out of project Athena. It really was a special time in compute and had some really novel concepts.<p>Source: I've spent waaaay too much time dealing with AFS and kerberos.
This kind of thing is so cool! I'm surprised some kind of distributed filesystems isn't already built into all the main OSes by now.<p>Like, 80% of desktop and mobile application features could be implemented with SyncThing as the backend.