A simulation study of the performance of multicast protocols

Abstract

We have chosen to study multicast protocols because of their applicability to a growing number of network applications. We have designed and implemented a simulation system to study multicast protocols without the need for a dedicated testbed network, and used the simulator to study two simple multicast protocols, the stop and wait and block acknowledgment protocols. We found that these protocols outperform equivalent unicast protocols for small number of receivers in error-free conditions. The block acknowledgment protocol shows higher throughput with larger window sizes, but has greater latency. Both protocols, however, deteriorate quickly as the number of receivers and failure rate of the underlying-network grow. This deterioration is caused by a rising collision rate among receivers sending acknowledgments which in turn aggravates network congestion. Throughput of both protocols is improved by using randomly timed acknowledgments to reduce the collision rate; however, this technique does not affect the high sensitivity of the protocols to error conditions.