A Resilient Transport System for Wireless Sensor Networks

A Resilient Transport System for Wireless Sensor Networks Chieh-Yih Wan This thesis contributes toward the design of a new resilient transport system for wireless sensor networks. Sensor networks have recently emerged as a vital new area in networking research, one that tightly marries sensing, computing, and wireless communications for the first time. Wireless sensors are embedded in the real world and interact closely with the physical environment in which they reside. These networks must be designed to effectively deal with the network’s dynamically changing resources, including available energy, bandwidth, processing power, node density, and connectivity. This dissertation focuses on making the sensor network transport system resilient to such changes in many cases abrupt changes. We define transport resilience as the ability of the network to deliver a sufficient amount of sensing events to meet the applications’ fidelity requirement for a set of different traffic classes while reducing the energy consumption of the network. More specifically, we investigate, study, and analyze two classes of transport resilience: (1) the need to reliably deliver data under various error conditions; and (2) the need to maintain the application’s fidelity under congested network conditions. We take an experimental systems research approach to the problem of supporting resilience in sensor networks by building an experimental sensor network testbed and evaluating a set of new resilient transport algorithms under various workloads and changing network conditions. We study the behavior of these algorithms under testbed conditions, and apply what is learned toward the construction of larger and more scalable resilient networks. This thesis makes a number of contributions. First, we propose a new reliable delivery transport paradigm for sensor networks called Pump Slowly Fetch Quickly (PSFQ). PSFQ represents a lightweight, scalable and robust transport protocol that is customizable to meet a wide variety of applications needs (e.g., re-programming, actuation, reliable event delivery). We present the design and implementation of PSFQ, and evaluate the protocol using the ns-2 simulator and an experimental wireless sensor testbed based on Berkeley motes and the TinyOS operating system. The PSFQ protocol represents the first reliable transport proposed for wireless sensors networks. Next, we present the design of an energy-efficient congestion control scheme for sensor networks called CODA (COngestion Detection and Avoidance). We define a new objective function for traffic control in sensor networks, which maximizes the operational lifetime of the network while delivering acceptable data fidelity to sensor network applications. CODA is founded on three important distributed control mechanisms: (1) an accurate and energy-efficient congestion detection scheme; (2) a hop-by-hop backpressure algorithm; and (3) a sink to multi-source regulation scheme. We evaluate a number of congestion scenarios and define new performance metrics that capture the impact of CODA on the sensing application performance. We analyze the performance benefits and practical engineering challenges of implementing CODA in an experimental sensor network motes testbed. CODA represents the first comprehensive solution to the congestion problem in sensor networks. The final contribution of this dissertation explores a complementary solution to CODA called dual radio virtual sinks that boosts the performance of sensor networks even under persistent overload conditions. We propose to randomly distribute a small number of all-wireless dual radio virtual sinks throughout the sensor field. In essence, these virtual sinks operate as safety valves in the sensor field by selectively siphoning off overload traffic in order to maintain the fidelity of the application signal delivered to the network’s physical sink. A key feature of virtual sinks is that they are equipped with a secondary higher bandwidth, long-range radio (e.g., the IEEE 802.11), in addition to their primary low bandwidth, low power mote radio. Virtual sinks are capable of dynamically forming a secondary ad hoc radio network that can be used on-demand by the mote radio network. Rather than ratecontrolling packets during periods of congestion (as is the case with CODA), virtual sinks take the congested traffic off the low-powered sensor network and move it on to the secondary radio network, transiting it to the network’s physical sink. We study, propose, and evaluate a set of algorithms for virtual sink discovery, selection, traffic transiting and load balancing. We leverage the use of the Stargate platform to support an all-wireless virtual sink approach in our sensor network testbed. We believe that sensor networks must be built to be robust to various software and hardware failures, and be resilient to dynamic resource changes such as node failures, increased packet error rates, and traffic surges. Collectively, PSFQ, CODA, and virtual sinks provide a set of energy-efficient, robust transport mechanisms that serve as a foundation for making sensor networks more resilient.