On the Convergence and Stability of Data-Driven Link Estimation and Routing in Sensor Networks

The wireless network community has become increasingly aware of the benefits of data-driven link estimation and routing as compared with beacon-based approaches, but the issue of Biased Link Sampling (BLS) estimation has not been well studied even though it affects routing convergence in the presence of network and environment dynamics. Focusing on traffic-induced dynamics, we examine the open, unexplored question of how serious the BLS issue is and how to effectively address it when the routing metric ETX is used. For a wide range of traffic patterns and network topologies and using both node-oriented and network-wide analysis and experimentation, we discover that the optimal routing structure remains quite stable even though the properties of individual links and routes vary significantly as traffic pattern changes. In cases where the optimal routing structure does change, data-driven link estimation and routing is either guaranteed to converge to the optimal structure or empirically shown to converge to a close-to-optimal structure. These findings provide the foundation for addressing the BLS issue in the presence of traffic-induced dynamics and suggest approaches other than existing ones. These findings also demonstrate that it is possible to maintain an optimal, stable routing structure despite the fact that the properties of individual links and paths vary in response to network dynamics.