A Recovery Protocol for Spritely NFS
Jeffrey C. Mogul · 1999
NFS suffers from its lack of an explicit cache-consistency protocol. The Spritely NFS experiment, which grafted Sprite’s cache-consistency protocol onto NFS, showed that this could improve NFS performance and consistency, but failed to address the issue of server crash recovery. Several crash recovery mechanisms have been implemented for use with network file systems, but most of these are too complex to fit easily into the NFS design. I propose a simple recovery protocol that requires almost no client-side support, and guarantees consistent behavior even if the network is partitioned. This proves that one need not endure a stateless protocol for the sake of a simple implementation. I also tidy up some loose ends that were not addressed in the original experiment, but which must be dealt with in a real system. This is a preprint of a paper presented at the USENIX Workshop on File Systems, May, 1992. Copyright 1992 Digital Equipment Corporation d i g i t a l Western Research Laboratory 250 University Avenue Palo Alto, California 94301 USA