Distributed File Systems
Stub. Extending the local file system abstraction across a network: the general distributed-systems concerns this raises, and two classic case studies — the Network File System (NFS) and the Andrew File System (AFS), which make very different tradeoffs on caching, consistency, and statefulness.
Once a file system’s storage lives on a different machine than its clients, every distributed-systems problem this repo already has notes on — caching, consistency, partial failure — shows up again in a concrete, very-widely-deployed setting. NFS and AFS are the two classic points of comparison: NFS’s stateless server design (simple, crash-recoverable, but weak client-side caching guarantees) versus AFS’s stateful, callback-based caching (better scalability and consistency, at the cost of server-side state to manage). See replicated-data-management-and-consistency for the general consistency-model vocabulary these two systems instantiate.
Outline (TODO — flesh out each)
- Distributed file systems — the problem, and what’s different from a local FS
- The Network File System (NFS)
- The Andrew File System (AFS)