Towards Super-Scalable Multicast

(This work is supported in part by
a National Science Foundation CAREER Grant (NCR-9624106))

Large-Scale multicast applications for the Internet require the availability of multicast protocols that can support multicast groups with many thousand simultaneous members. For this it is crucial that there exist mechanisms to efficiently exchange control information between the members of a group. In this paper, we present a new approach for distributing control information within a multicast group which increases the scalability of multicast applications. Multicast group members are organized as a logical {\em n-dimensional hypercube},and all control information is transmitted along the edges of the hypercube. We analyze the scalability of the hypercube control topology and compare it with tree-based approaches. We show that the hypercube balances the load per member for processing control information better than existing topologies. We use actual data traces of the group membership in an MBONE conference to gain insight into the transient changes of the load at each node. In a subsequent companion technical report \cite{LiSe98b} we will present a set of soft-state protocol mechanisms that maintain the hypercube topology without requiring any entity to have global state information.