Traditional ARQ-based reliable protocols for unicast (e.g., TCP) as well as multicast (e.g., Horus, RMTP etc.) use sequential numbering of data units and detect losses from discontinuities in the sequence of received packets. The Application Level Framing (ALF) model encourages application control over loss-detection and recovery. With sequence numbers, the application must express its reliability requirements using sub-sequences of the sequence space. This is both cumbersome and restrictive for applications that have no a priori knowledge of the data stream. Distributed whiteboard applications, webcast, and file system multicasting are some examples of applications where data is continuously generated and receivers cannot predict which sub-sequences must be received reliably.

In this paper, we propose an alternative data naming scheme which enhances the expressibility of applications' reliability and ordering requirements. We apply the new data naming scheme to build a framework for light-weight Scalable, Reliable Multicast (SRM) sessions, and develop a State Announcement Protocol (SAP) for loss detection. Using simulations, we study the scaling behavior of SAP and show how the protocol scales well for large group sizes. We also suggest heuristics for certain classes of applications that improve the convergence times and message complexity of the protocol.




Download Full History