A replicated Unix file system
Barbara H. Liskov, Robert E. Gruber, Paul Johnson, Liuba Shrira · 1990
The Programming Methodology group at the Laboratory for Computer Science at M.I.T. is working on an implementation of a replicated Unix file system for use via the NFS protocol [10,12].Our replication method is intended to support the following goals:1.When used via NFS, our system should provide the same semantics as an unreplicated NFS server.Furthermore, it should be usable with whatever NFS client code exists at the client machine. The system should not depend on proprietary information.3. The system should continue to provide service even when one replica is crashed or inaccessible, but have only two copies of each file.4. The system should perform as reliably as a single, unreplicated NFS server.5. The system should provide response time comparable to that provided by a single NFS server.In particular, the delay observed by the client in doing a read or write should be no greater with our service than with a single server.Our work is most similar in its goals to [6], but we use a general replication algorithm tailored to this application rather than building on top of an existing replication technique; we expect to achieve better performance in this way.We differ from the work on Coda [11] because we take the NFS/Unix file system semantics as a given, and because we are not willing to implement front ends on client machines.We also differ from the work on Echo [5] for these reasons, although Echo will support Unix file system semantics as an option, and their replication technique is similar to ours.