On the Correlation between Spectral Measures of Network Topology and Flow-Level QoS

Abstract

In recent years, the spectrum of a graph, which is derived from the properties of eigenvalues and eigenvectors of the matrix corresponding to the graph has attracted attention. Since the topology of a communication network can be represented as a graph, analyses of the relationship between spectral measures of the network topology of a communication network and the characteristics of the communication network have been performed. In this paper, we experimentally investigate the relationship between spectrum measures of a network topology and flow level QoS (Quality of Services) of the transport protocol operating on the network. We aim to answer the following research question: from spectral measures of the network topology of a communication network (i.e., the spectral radius, the spectral gap, the natural connectivity, the algebraic connectivity, and the effective graph resistance), how well can we estimate the QoS (i.e., throughput, round-trip time, and packet loss rate) of dynamic flows operating on the communication network? Our findings include that among five spectral measures, the effective graph resistance is most suitable for estimating the flow-level Qos of TCP flow.