Tree-Based Fault-Tolerant Multicast in Multicomputer Networks Using Pipelined Circuit Switching

A tree-based fault-tolerant multicast algorithm built on top of pipelined circuit switching is presented. For every multicast message, a multicast tree is constructed in a distributed and adaptive fashion. An underlying fault-tolerant routing algorithm is used to tolerate faulty nodes and links without requiring nodes to have global fault information. The multicast algorithm is provably deadlock-free and livelock-free, requires only a single message startup, and guarantees all messages are delivered over shortest paths in the fault-free and traac-free case. Simulation results show that the algorithm produces shorter average communication latency than previous path-based fault-tolerant multicast algorithms over a range of network loads and fault conditions.