Simplification of Network Dynamics in Large Systems

Xiaojun Lin, Purdue University

Abstract

Controlling today's communication networks is a challenging task due to the tremendous growth both in terms of the network capacity and in the number of elements (e.g., end-users, routers and hosts) that the network supports. As the size of the network grows, the network dynamics also become increasingly complex. Solutions that once work well for small networks may no longer be appropriate for large-scale and high-bandwidth networks. In this dissertation, we have taken two orthogonal approaches to simplify the network dynamics in large communication systems. In the first approach, we seek simplicity through exploiting the largeness of the network. We first study the pricing-based control problem and the Quality-of-Service routing problem in large-capacity wire-line networks. We show that simple static control policies can approach the performance of the optimal (but complex) dynamic control policy when the capacity of the system is large. We develop simple and distributed control algorithms based on these static control policies. Our control solution can significantly reduce the computation complexity and communication overhead without sacrificing the network performance. We then turn to wireless networks and investigate the fundamental tradeoff between the capacity and the delay in large mobile wireless networks. By exploiting the largeness in the number of nodes in these networks, we obtain simple scaling laws that determine the optimal achievable capacity given delay constraints. In the second approach, we seek simplicity by designing an appropriate control architecture such that complex interactions within the system can be structured into layers that are only weakly dependent on each other through a judiciously chosen set of control parameters. In particular, we investigate the cross-layer congestion control and scheduling problem in multi-hop wireless networks. We develop a loose-coupling approach to this problem, where the cross-layer solution only requires a minimal amount of interaction between the layers, and is robust to imperfect decisions at each layer. This result allows us to use imperfect, but simpler and potentially distributed, algorithms for cross-layer control of large wireless networks. We have successfully developed such a fully distributed cross-layer control solution for certain interference model.

Degree

Ph.D.

Advisors

Shroff, Purdue University.

Subject Area

Electrical engineering

COinS