Delay considerations in multicast routing for ATM networks

Many of the services envisaged for Broadband ISDN must be capable of operating on a point-to multipoint basis. Examples are video distribution, multi-media conferencing and broadcast information services. To support these services, the network must be capable of handling multicast virtual channels. Multicast facilities are included in the majority of fast packet switch designs for ATM networks, but less work has been done on the routing of multicast transmissions through networks of arbitrary topology connecting such switches. There are two principal requirements for multicast routing in a high-speed network. First, the network should be used efficiently e.g. capacity that can be shared should be shared. Secondly, because of the real time aspect of many multipoint services, it is important to minimise delay to the recipients. The first topic has been addressed by authors such as Waxman, who examines heuristic techniques for reaching an approximation of a Steiner tree solution. We have also looked at the convergence properties of a number of iterative solutions to this problem based on a minimum spanning tree. However, in this paper we concentrate on the delay which we see as a factor of equal, if not greater, importance. The main elements that contribute to the delay through an ATM network are the fixed switching delay at the nodes, the propagation delay experienced between nodes and the queuing delay encountered for the inter-node links. To minimise the propagation delay, a shortest path (SP) routing strategy is appropriate from which a broadcast SP can be built with a delay bound equal to the longest delay from the source. We show that, where link costs are proportional to delay costs, a modification to Dijkstra's SP algorithm can give a more efficient solution than the original Dijkstra algorithm. Results will be presented for both randomly generated networks disregarding delays at the node and for simulated ATM networks.