File:Traffic congestion analysis for a software-defined network (IA trafficcongestio1094558337).pdf

The objective of this thesis is to implement an anomaly-detection method that can be used to detect congestion in a software-defined network. The method incorporates spectral graph theory and phantom node techniques. The experimental implementation of spectral graph theory used eigenvalue-eigenvector solutions to characterize a mathematical model of the network’s topology. In this thesis, we used the phantom node technique to determine congestion in the network by using a virtual node to set the threshold for available link capacity, or the maximum amount of traffic, that can cross the links in the network before the links are considered congested. Results show that when the network is congested, a shift occurs in the eigenvalue and eigenvalue index spectrum. Prior to congestion, the virtual node has the highest nodal influence in the lowest eigenvalue index; however, when a node becomes congested and high traffic in the node crosses the threshold set by the virtual node, the congested node takes the position of the virtual node in the eigenvalue index. The virtual node shifts to having the greatest nodal influence in the next-higher eigenvalue index in the spectrum. Essentially, the results show that anomalies, such as congestion, can be detected using the anomaly-detection method developed in thesis.