Design and evaluation of network reconfiguration protocols for mostly-off sensor networks

A new class of sensor network applications is mostly-off. Exemplified by Intel’s FabApp, in these applications the network alternates between being off for hours or weeks, then activating to collect data for a few minutes. While configuration of traditional sensornet applications is occasional and so need not be optimized, these applications may spend half their active time in reconfiguration every time when they wake up. Therefore, new approaches are required to efficiently “resume” a sensor network that has been “suspended” for long time. This paper focuses on the key question of when the network can determine that all nodes are awake and ready to communicate. Existing approaches assume worst-case clock drift, and so must conservatively wait for minutes before starting an application. We propose two reconfiguration protocols to largely reduce the energy cost during the process. The first approach is low-power listening with flooding, where the network restarts quickly by flooding a control message as soon as the first node determines that the whole network is up. The second protocol uses local update with suppression, where nodes only notify their one-hop neighbors, avoiding the cost of flooding. Both protocols are fully distributed algorithms. Through analysis, simulation and testbed experiments, we show that both protocols are more energy efficient than current approaches. Flooding works best in sparse networks with six neighbors or less, while local update with suppression works best in dense networks (more than six neighbors).