Fun exercise! 1) The old school UNIX admin solution to this is to make each of your fileservers NFS servers as well and cross-mount them all with each other.

Fun exercise! 1) The old school UNIX admin solution to this is to make each of your fileservers NFS servers as well and cross-mount them all with each other. Pros: it's there, easy, and pretty simple to understand Cons: NFS clients (the whole OS) can get cranky when a NFS server disappears silently and the client attempts to access a file on it. 2) You can set up a machine with small, fast disk, and a reverse-proxy caching solution. Then you can point that cache server at a single server with acres of cheap disk. Pros: This can be more economical than scaling machines if you're careful: buying SATA ports and connecting more disks can be cheaper than buying whole computers (and rack space, and power, and switch ports) for their SATA ports. Cons: This technically doesn't solve the problem that you're eventually bounded by a single system's capacity. Also for certain pathologically bad workloads the cache server can thrash and performance will fall off.