Brad Rosen

CS 490

Advisor: Yang Richard Yang

Background

This work was inspired by the works of Rao, Shenker, Lin and Govindan. From Rao and Shenker, the merged concepts of radio topologies vs physical topologies and their impact on virtual topologies and positioning combined with the rendezvous algorithm from Lin lead to the question of the effectiveness of directed mobility. As we can see, the real significance lies in the radio graph, as is reflected by varying alphas in the directed mobility scheme.

Abstract

In the near future, the advent of large-scale networks of mobile agents autonomously performing long-term sensing and communication tasks will be upon us. However, using controlled node mobility to improve communication performance is a capability that the mobile networking community has not yet investigated. In this paper, we study mobility as a network control primitive. More specifically, we present the first mobility control scheme for improving communication performance in such networks. Our scheme is completely distributed, requiring each node to possess only local information. Our scheme is self-adaptive, being able to transparently encompass several modes of operation, each respectively improving power efficiency for one unicast flow, multiple unicast flows, and many-to-one concast flows. We provide extensive evaluations on the feasibility of mobility control, showing that controlled mobility can improve network performance in many scenarios. This work constitutes a novel application of distributed control to networking in which underlying network communication serves as input to local control rules that guide the system toward a global objective.

Methodolgy

A simulator, written from scratch in C#, was used to model arbitrary topologies to see if nodes would yield a power savings when given the opportunity to move to save power. Both constrained and unconstrained cases [preserving and destructive of network connectivity, respectively] were examined.

PDF [Complete Paper]

Simulator Source Code [Finalized as of Tuesday, Dec 10^th]