Peer-to-Peer Storage Systems

Peer-to-peer storage systems such as TotalRecall and Om could also benefit from the efficiencies of us- ing FUSE to implement liveness checking. For example, TotalRecall relies on the overlay for liveness-checking of eager replicas, but must separately implement liveness- checking of lazy replicas. The substitution of FUSE groups would be straightforward. Om implements its own failure detection and timeout scheme using leases; these leases could be replaced by FUSE groups. FUSE would also be a good fit for Om because every replica in Om can regenerate the entire replica set, and therefore should monitor the liveness of all other replicas. This symmet- ric responsibility exactly corresponds to the semantics of FUSE group notifications. Lastly, the potential for false positives using FUSE does not compromise Om’s consistency guarantees; Om’s failure-induced reconfigu- ration protocol is already designed to be robust to failure- detection false positives.

In addition, FUSE may also be useful in some of the application areas targeted by weakly consistent member- ship services. For example, Vogels and Re argue that weakly consistent membership services would ben- efit many Web Services, ranging from scientific comput- ing to federated business activities. FUSE may be a more suitable choice for some of these emerging applications.