A Distributed Fault-Detection and Recovery Protocol for Reliable Multicast Collaborative Communications

Reliable multicast transport protocols support dissemination communication and their through-puts are either limited by the sender or by intermediate nodes that consolidate acknowledgements and retransmissions. RMSP is a portable reliable multicast protocol that supports collaborative communication at the session-layer level and provides higher throughput by maintaining two transport connections per node irrespective of the number of nodes. However, the throughput only of RMSP may degrade if one or more nodes or the centralized connection manager that monitors them fails. In this report, we propose a fully distributed fault-tolerant protocol, DFT-RMSP, that detects and isolates faulty nodes and reinstates the multicast collaborative communications. We derive analytic expressions for throughput degradation and analyze its sensitivity to node-failure rates, network delay, and multicast group size. We show that DFT-RMSP provides less down-time and achieves graceful throughput degradation under high failure rates compared with RMSP. simultaneous node failures.