LRRM: a randomized reliable multicast protocol for optimizing recovery latency and buffer utilization

An efficient recovery protocol for lost messages is crucial for supporting reliable multicasting. The tree-based recovery protocols group nodes into recovery regions and designate a recovery node per region for buffering and retransmitting lost messages. In these protocols, the recovery host may get overloaded during periods of large message losses and costly remote recovery may be initiated even though a peer node has the lost message. To address these drawbacks, the randomized reliable multicast protocol (RRMP) was proposed which distributes the responsibility of error recovery among all members in a group. The pressure on the buffer and computational resources on the intermediate nodes is increasing due to the wide distribution of multicast participants with widely varying reception rates and periodic disconnections. In this paper, we propose the lightweight randomized reliable multicast (LRRM) protocol that optimizes the amount of buffer space by providing an efficient mechanism based on best-effort multicast for retrieving a lost message. A theoretical analysis and a simulation based study of two realistic topologies indicate that LRRM provides comparable recovery latency to RRMP for lower buffer space usage. While presented in the context of RRMP, LRRM can also benefit other tree-based reliable multicast protocols.