A SIMPLE ADAPTIVE ROUTING SCHEME FOR CONGESTION CONTROL IN SHUFFLENET MULTIHOP LIGHTWAVE NETWORKS

Under uniform traffic conditions, a ShuffleNet multihop lightwave network makes efficient use of the optical channels with a simple fixed routing algorithm. In this paper, we describe a simple adaptive routing scheme for datagram (connectionless) and virtual circuit (connection-oriented) transmission that relieves congestion resulting from nonuniform traffic patterns and network failures. The ideas presented here may be used for other multihop lightwave networks. We focus on ShuffleNet because its "tree structure" can quickly disperse packets away from congested portions of the network. The distributed routing algorithm uses only local state information available at the users, and does not require a priori knowledge of the traffic patterns. It delivers packets in the minimum number of hops when the network is "congestion free," but as a channel becomes overloaded it distributes some of the traffic over less busy channels by automatically "bumping" packets to longer routing paths. In contrast with typical deflection- or hot-potato-type routing algorithms, a packet is bumped at most once, thereby yielding a bounded end-to-end delay with a small variance. Moreover, the adaptive algorithm decreases the chance for congestion to develop (and the need to bump packets) by routing each packet over less busy channels whenever there exist multiple minimum-hop routing paths to the packet's destination. Simulations of the adaptive routing scheme for datagram transmission demonstrate its ability to support nonuniform traffic patterns, reduce the mean and variance of the queue sizes, and require small resequencing buffers. A virtual circuit version of the adaptive routing algorithm eliminates the need for resequencing buffers.