Scalable Reliable Multicast with Hierarchy and Polling

The IP multicast architecture enabled large-scale applications of multicasting on the Internet. Many of these applications require reliable dissemination of a stream of data (e.g., software distribution or stock updates) to a large number of receivers. They face, however, a scaling limitation, known as the feedback implosion problem: the sender is overwhelmed by feedback packets, leading to packet losses. Recent reliable multicast protocols have been designed with scalability in mind. They usually follow a receiver-initiated approach: the sender is unaware of receivers; it transmits to a group id without waiting for feedback and handles retransmission requests when possible. Receiver-initiated schemes trade in eeciency and reliability for improved scalability. This paper describes an alternative approach to scalable reliable multicasting: instead of resorting to receiver-initiated schemes, the scalability of sender-initiated protocols is enhanced through polling feedback and hierarchy. A novel polling-based implosion avoidance mechanism reduces the amount of feedback packets to desired levels, and thus avoids implo-sion, while the hierarchical organization is harnessed for increased scalability, local recovery, as well as improved ow and congestion control. The resulting protocol is called prmp: polling-based reliable multicast protocol.