Deterministic and adaptive routing policies in packet-switched computer networks

Abstract

Efficient routing policies are required for both design and operation of distributed, packet-switched computer networks. In the network design process, deterministic policies, which are time invariant, easy to analyze and simple to represent, are generally used. For the routing of packets in a real network, on the other hand, adaptive policies, which can adjust to traffic fluctuations and component failures are implemented. Therefore, a computer network is typically optimized with respect to a deterministic policy and then operated with an adaptive policy.In the first part of this paper the two policies are compared using analytical and simulation methods, and it is shown that, in uniform traffic situations, the flow patterns generated are very similar. This result justifies the use of deterministic policy in the design phase.In the second part of the paper, the inadequacy of traditional adaptive policies in handling particular traffic situations is pointed out. As a solution, a new centralized adaptive policy is proposed, which combines the positive features of both deterministic and adaptive routing. The feasibility of such centralized policy implementation on a real network, and its efficiency with respect to the traditional implementation is discussed.