UNICAST MESSAGE ROUTING IN С OMMUNICATION NETWORKS WITH IRREGULAR TOPOLOGY

In this paper we consider the problem of deadlock-free unicast wormhole routing in computer and communication networks with irregular topologies. An example of such networks are Network of Workstations (NOWs). In general, the topology of these networks can be quite random. Several methods exist in the literature for wormhole routing in networks/multiprocessors with a regular topology, such as a n-dimensional mesh, but very few papers have been published on wormhole routing for irregular networks. Some of these existing techniques require complex signaling hardware at the routers or result in a large amount of congestion at some specific links. The problem of deadlock-free routing consists of two parts. First, all deadlocks must be eliminated. An usual way of doing this, both for regular and irregular topologies, is to forbid some turns. The second part, which is the focus of this paper, is the problem of selecting an optimal (usually the shortest) path after the restrictions on routing have been formulated. We propose three efficient approaches for solving this problem. These approaches (local, global and mixed) differ in a way distances in the network graph are estimated using local information stored in the routers. Our approach for non-adaptive unicast deadlockfree wormhole routing provides for message paths very close to the shortest ones and more uniform distribution of the traffic between communication links in the system. Initial simulation results presented in the paper indicate that the proposed approaches are promising in terms of both throughput and scalability.