The neural network approach to a parallel decentralized network routing.

Abstract

With the progress of high-speed optical transmission and packet switching, a large capacity packet-based multi-media communication network is expected to spread rapidly. One of the key issues in these networks is the network routing that chooses the route to the destination for packet transmission in the network. In most previous work, the whole network is mapped to a large size-Hopfield-type neural network. Hence, the network routing by this method is not beyond the centralized control. In this paper, a parallel decentralized Network Routing method is presented. The model comprises an interconnection of groups of an intraconnected network, which is fully connected, and resides at each node of the communication network. Since the dynamics of each neuron in the whole system follows a unique state equation, we can see easily how the update of a neuron maps to real world network routing problems. Most important, becauase of the dynamics of the neurons with such a high speed of convergence, the model has the ability to achieve a sub-optimum routing solution in a real-time application. Finally, simulation results validate the proposed method.