Router-level topology modelling and assessment

Abstract

Modelling the Internet router-level topology is very important for network designers and administrators. A realistic router-level model make it possible to carry out performance evaluation, and enables the prediction of future traffic given some previous traffic data or models. In this thesis, our research on the router-level modelling issues and assessment methodology are presented. We first introduce several popular models that have important influence on the development of complex networks. Then, we build our test router-level networks by integrating the layered modelling approach (heuristically optimal topology) and scale-free growth algorithm (preferential attachment). The theoretic evaluation shows that different types of routers could lead to different evolution patterns of network throughputs and router utilizations. Secondly, we study the traffic generation issue and propose an efficient framework for traffic generation in order to investigate the scaling properties of the router-level topologies; constrained by technological and economic factors, and guided by empirical data. A more realistic evaluation result is obtained and some potential congestion areas are located. Finally, we explore the essences of layered modelling approach as an appropriate router-level modelling method. Some meaningful insights to the router-level modelling are also provided to inspire the future directions of work.