Characterization and performance analysis of a cognitive routing scheme for a metropolitan-area sensor network

Abstract

This MEng thesis is an exploration of the notion of cognitive methods for routing in a network, and the resulting potential for improvements in network performance. In cognitive routing, individual network nodes gain information about the state of the network in a distributed fashion, by measuring observable data such as packet arrival counts and timing. The nodes then use inference and estimation methods on the network traffic to modify the parameters of their routing protocols and/or routing tables, in order to improve some performance metric such as packet delay or network throughput. In this project we provide an example of the performance improvements possible through cognitive routing, by demonstrating a simple but nontrivial use of network measurement and inference to minimize the maximum average packet delay, and increase the max load that the network can handle. With more information-rich metrics that are available to be passively gathered by a routing protocol, such as source-destination IDs, the sizes of packets passing through a node, and packet loss rates, cognitive routing protocols may be able to predict congestion or link failures, potentially leading to much greater efficiency gains than are described in this project.