Study of communication network performance from a complex network perspective

Abstract

This thesis studies the performance of communication networks from a network science perspective. Our task is to establish a clear link between some structural properties of networks, such as degree distribution, average distance between nodes, and betweenness, with communication network performance, the purpose being to improve understanding of the various factors that affect the performance of communication networks and to provide design information for optimizing performance. Communication networks are modeled, analyzed and characterized using complex network parameters. In particular, the impact of network topology, routing strategy and resource allocation on the performance of generic communication networks is studied through theoretical analysis and computer simulation. Specifically, the regular lattice, ER random, BA scale-free and Internet AS-level networks under shortest-path (SP) and minimum-degree (MD) routing strategies with various types of resource allocation schemes are considered for determination of network parameters. Performance parameters, including packet drop rate, time delay, and critical generation rate, are evaluated. Node usage probability is proposed as a new metric for characterizing the traffic load distribution and how frequently a node is chosen to relay packets in a network. Based on the concept of node usage probability, effective network design strategies, including routing algorithms and resource allocation schemes, can be developed to maintain balanced traffic loads in the network nodes by avoiding overuse of certain nodes. The performance of the proposed minimum-node-usage routing strategy is compared with that based on other popular routing algorithms. Moreover, the effects of different types of traffic generation sources on network performance are studied. Finally, network design strategies for optimizing the performance of communication networks will be proposed. For efficient and reliable data transmission, the traffic load should be as uniformly distributed as possible in the network and the average distance travelled by the data should be short. This criterion has been shown to be fundamental. The key design problem is therefore to find the optimal solution that achieves this criterion. With a fixed network topology, the traffic load distribution and the transmission efficiency are determined by the specific routing algorithm and the traffic generation pattern. Specifically, a simulated annealing algorithm is employed to find the near optimal configuration of network design, which effectively balances traffic loads and improves the overall traffic performance.