Abstract

A novel highly scalable loss recovery algorithm for Multicast transmissions in active networks that achieves near-optimal implosion and very low latency is proposed. Quasi-minimum implosion is attained by stochastically selecting a sub-set of the loss-affected members as NACK senders. If appropriately tuned, the algorithm selects with high probability only one member as a NACK sender. Near-optimal latency is obtained by minimizing the time taken to select the NACK sender and by retransmitting from the closest possible location. Performance evaluation results show a maximum implosion 4% higher than the optimal value and a low latency. © 2007 International Federation for Information Processing.