The Effect of Link Churn on Wireless Routing

In this paper, we examine the spatiotemporal dynamics of wireless links and evaluate their effect on routing. We ran several experiments using a testbed consisting of 57 MicaZ motes, and collected data on link behavior over one entire day. We use this data to observe the overall network connectivity over time and space. We are able to examine in detail the choice of neighbors and routes using several linkselection mechanisms, both statically and over time. We are able to verify the hairy-edge hypothesis, which states that the most important links for routing are the most difficult to predict. In order to do so, we develop precise definitions of important and unpredictable links. We also find it possible to remove these intermediate links from consideration and still have a very rich set of links to route over, while suffering from fewer difficult-to-predict links. Also, we explore the tradeoff between statically defining routes as opposed to a dynamic protocol. We find that while it is not possible to remove all temporal variations from the network, their impact can be significantly reduced through the use of local redundancy. Finally, we present a survey of how several existing routing protocols fit into the framework developed in the body of the paper.