A distributed optimal routing strategy based on an estimation of the OD matrix

Abstract

This paper presents a distributed optimal routing strategy that is based on an estimation of the OD (origin-destination) matrix. This routing strategy may be employed in packet-switched networks. In the most common form of distributed routing, every node acts autonomously, distributing its link delay information to all other network nodes and periodic or aperiodic computing its own routing table based upon the link delays information obtained from other network nodes. The optimal routing assumes that the OD matrix is known. In order to avoid the transportation of the OD matrix through the communication network, it is estimated based on the link delays, the transit flows and the prior OD matrix. The represented routing strategy involves a computationally intensive iterative procedure, and therefore within large networks, this routing strategy becomes impractical. In these networks, during normal operation of the network, without changes in the network topology, a primitive routing algorithm such as a distributed Frank-Wolfe with fixed stepsize might be used. In addition, every node estimates the OD matrix. Then in the case of changes in the network topology, the estimated OD matrix is used to compute the routing tables. A further application of this strategy is within the network performance monitoring.