Recovery Timer Adaptation in SRM

SRM is a generic framework for reliable multicast data delivery. The retransmission requests and replies used in SRM error recovery are multicast to the entire group. SRM uses random timers to avoid the message implosion problem, and the eeectiveness of SRM's duplicate suppression depends critically on the design of these random timers. In order to achieve good performance in a dynamic network environment, the random timers should be adaptive to the network and session dynamics. This paper focuses on the scaling behavior of the random timer adaptation mechanisms in SRM. Through analysis and simulations, we analyze the relationship between the timer setting parameters and performance, investigate the representative timer adaptation approach proposed in 1], and present a collaborative approach of random timer adjustment. Generally speaking, the representative approach elects representatives to send retransmission requests or replies; whereas the collaborative approach distributes the error recovery workload among members.