Tree-based Reliable Multicast Protocols Research Articles

Comparison of tree-based reliable multicast protocols for many-to-many sessions

Tree-based reliable multicast protocols are well-known for being scalable for one-to-many sessions, but their behaviour in terms of throughput and bandwidth requirements has not been identifed for many-to-many sessions. In this paper, existing tree-based reliable multicast protocols are categorised into four classes according to their feedback/retransmission mechanisms. Analysing and comparing them using a new spatial multicast loss model that considers the correlation of packet loss events can yield realistic analysis results. The analysis reveals that for many-to-many sessions, tree-based protocols with unicast NACK, unicast retransmission, and periodic polling are most scalable with respect to throughput and network bandwidth for practical session sizes. The result is contrary to the one-to-many case and it is shown that the reason is that reducing the sender's processing time is a key factor for scaling one-to-many reliable multicast, whereas reducing the receiver's processing time is far more important factor for scalability of many-to-many reliable multicasts.

Optimal branching factor for tree-based reliable multicast protocols

In recent years, many reliable multicast protocols on transport layer have been proposed. Previous analysis and simulation studies gave evidence for the superiority of tree-based approaches in terms of throughput and bandwidth requirements. In many tree-based protocols, the nodes of the tree are formed of multicast group members. In this case, the branching factor, i.e. the maximum number of child nodes is adjustable. In this paper, we analyze the influence of the branching factor on a protocol's throughput and bandwidth consumption. This knowledge is important to configure protocols for best performance and to optimize the tree creation process. Our results show that the optimal branching factor depends mainly on the probability for receiving messages from other local groups. If local groups are assigned to a separate multicast address, the optimal branching factor is small. On the other hand, if TTL scoping is used and therefore the probability for receiving messages from other local groups is greater than zero, larger local groups provide better performance.