A Fuzzy System for Adaptive Network Routing

In this paper we propose an adaptive routing algorithm in which the link cost are dynamically assigned using a fuzzy system. The traffic in the network is re-routed to nodes, which are less congested, or have spare capacity. Based on a set of fuzzy rules, link cost is dynamically assigned depending upon the present condition of the network. Distance vector algorithm, which is one of the shortest path routing algorithms is used to build the routing tables at each node in the network. The proposed fuzzy system determines the goodness of a link given the present congestion situation measured via the delays experienced in the network and the offered load on the network. Delay in the links, was estimated by the time taken for the test packets to travel from the node to its neighbors. The information collected by the test packets and the number of packets waiting in the queue, are the two inputs to the fuzzy system. The output of the fuzzy system is cost of the link for a certain interval. This algorithm was applied on a simulated NSFNET, the USA backbone as well as to another test network with a different topology. Robustness and optimality of the algorithm was tested by simulating various types of load patterns and by comparing with some of the existing algorithms. The proposed fuzzy based algorithm always gave optimal performance under varying load conditions and topologies.