A two phase multicast routing algorithm in self-routing multistage networks

We present an improved multicast routing algorithm for self-routing multistage networks with circulating data. This multicast routing algorithm uses two passes through the network, and is based on the cube concept where a cube consists of a group of outlets reachable in one pass through the multistage network. A given multicast connection set, consisting of D destinations, is decomposed into a set of m cubes so that each cube can be routed in one pass through the multistage network. In the first phase of the multicast routing algorithm, data is routed from the source to m consecutive outputs. These m outlets are then used as inputs in the second phase for routing to the destination cubes. When the cubes satisfy a certain condition, we show a method used to map each of the m consecutive inputs to route data to one of the cubes of the multicast connection. When some cubes conflict and do not satisfy this condition, we can still route in two passes by expanding a few larger cubes into smaller cubes in order to satisfy this condition for routability. We evaluate the performance of this two-phase algorithms in terms of number of internal links used via extensive simulation. This multicast algorithm using the cubes performs significantly better than the simple multicast algorithm which routes to D consecutive outputs in the first phase and to final destinations in the second phase.