FUSE Implementation
Our FUSE implementation is scalable, where scaling is with respect to the number of groups: multiple failure notification groups can share liveness checking messages. Other implementations of the FUSE abstraction may sac- rifice scalability in favor of increased security. Our FUSE implementation is designed to support large numbers of
small to moderate size groups. We do not attempt to effi- ciently support very large groups because we believe very large groups will tend to suffer from too-frequent failure notification, making them less useful.
Our FUSE implementation is particularly well-suited to applications using scalable overlay networks. Scalable overlay networks already do liveness checking to main- tain their routing tables and FUSE can re-use this live- ness checking traffic. In such a deployment, the network traffic required to implement FUSE is independent of the number of groups until a failure occurs; only creation and teardown of a group introduce a per-group overhead. FUSE can be implemented in the absence of a pre-existing overlay network – the implementation can construct its own overlay, or it can use an alternative liveness checking topology.
