On the Cons truc t ion of Low Cost Mul t icas

This paper addresses the problem of routing many-to-many multicast connections with bandwidth reservation. We devise a routing algorithm which provides not only connectivity but also bandwidth among multicast members at a low cost. After briefly addressing the problem complexity (NP-Complete), we move on to design polynomial time algorithms, for scalability purposes, some of which with worst case bounds on performance. A simulation study is used to illustrate the algorithms' performance and trade offs. 1 I n t r o d u c t i o n In a QOS environment, routing algorithms must provide routes which conform with users end-to-end requirements while minimizing the cost of a connection. A connection cost is defined as the sum of the costs of individual links participating in the connection. Link costs are assigned so as to reflect the desirability of using a given link as part of a connection path. QOS requirements translate into path constraints for a routing algorithm, and impact its computation complexity. Complexity is a major concern in network scalability since the routing algorithm must sometimes run in real time, specially in case of multicast connections. If route computation takes too long, topology information in which this computation is performed is more likely to be stalled, leading to a higher probability of having a connection setup failure. The first, and perhaps most important network resource is bandwidth. We assume a link state routing strategy, as in PNNI ATM technology or in Internet OSPF used as a Interior Gateway routing Protocol (IGP), where each route computation site knows about resources along the links of the entire network. The problem of routing unicast (one-to-one) unidirectional connections with bandwidth reservation was shown to be solved by a Dijkstra type algorithm [Wang95] for link state routing. For many-to-many type of connections, however, to our knowledge there is no routing algorithm in the literature which includes bandwidth reservation. This work was partially supported by CNPq under Grant 201597/93-4(NV), and by GTE, NSF, and MICRO-SUN