This project takes a fresh look at challenges in sensor networks in light of recent technology trends and experiences in pilot deployments. Technology trends indicate that the capacities of flash memories will continue to rise while their costs and energy consumption continue to plummet. This will make it possible to equip sensor nodes with energy-efficient, high-capacity NAND flash memory storage. Pilot deployments have shown that scalable sensor network architectures will be hierarchical, and comprise several hundreds of resource-constrained sensors as well as several tens of resource-rich sensor proxies. This motivates the need to develop novel methods to exploit resources at proxies while respecting constraints at sensors. This project has two contributions. The first contribution is a system for storage and retrieval of archival sensor data. This research includes the design, prototyping and evaluation of archival storage subsystems for sensor nodes, algorithms to enable efficient access of large distributed archival sensor data, and compression techniques for efficiently retrieving such data. Second, this project proposes an uncertainty-driven energy management architecture that unifies energy optimization across sensing, communication, routing, data processing and query processing tasks. This research uses prediction models and uncertainty as fundamental building blocks and builds a spectrum of energy-optimized services over this foundation. The project will have broad impact across data-intensive sensor network applications in science and engineering, as well as on education across the Five College consortium. The results of this project including publications, software and hardware prototypes will be made freely available to the research community.
The dramatic growth of sensor networks has seen the emergence of a variety of applications and deployments that span scientific, engineering, medical and other disciplines. This proposal addresses two important research problems in sensor networks: data management and energy management. Our first contribution is the design and development of an archival storage system for sensor systems that exploits technology trends in flash memories. Our work showed that NAND flash storage is many orders of magnitude cheaper energy-wise than communication and comparable to the cost of computation on Mote-class sensor platforms. This study laid the foundation for several systems and networking modules that took advantage of low-power storage. 1) We designed Capsule, an energy-efficient flash-based storage substrate for sensor platforms that provides the abstraction of typed storage objects to applications. 2) We designed a utility-driven archived sensor data retrieval engine that served multiple concurrent users such as meteorologists who provide weather forecasts, scientists conducting research using "real-world" data, automated weather monitoring applications, and emergency response managers, 3) We designed a data retrieval system that can be used for archival image search and retrieval from a distributed camera sensor network, and 4) We designed a transport protocol that can take advantage of the significant flash storage capacity of sensor or mesh devices. Our second contribution is an uncertainty-driven energy management architecture that unifies energy optimization across sensing, communication, routing, data processing and query processing tasks. This work uses prediction models and uncertainty as fundamental building blocks and builds a spectrum of energy-optimized services over this foundation. Our work has led to 1) the PRESTO data management system that exploit phenomena models to reduce the overhead of querying sensor data, 2) adaptive duty-cycling techniques for energy harvesting-based sensor networks that are dynamic to energy variations, and 3) an uncertainty-driven routing protocol that is robust to dynamics of wireless links. We evaluated and demonstrated the benefits of our system across different applications. One application was activity monitoring of the elderly using low-power cameras. We addressed the problems of detecting when elderly have fallen and notifying medical staff. We also explored the problem of object finding, i.e. locating commonly misplaced objects such as keys, cellphones, PDAs, and books by elderly users. We also evaluated our work in the context of a severe storm monitoring sensor network being deployed by the Center for Adaptive Sensing of the Atmosphere at UMass (CASA). The PI also introduced a new graduate course on sensor networks, which has been taught since 2005. This course introduced students to both theoretical and practical aspects of sensor networks, including several research themes that were developed using the funding from this proposal including: a) development of storage systems for flash memory, b) design of camera sensor network testbeds, and c) data compression and communication in sensor networks. The course also includes a semester long project which exposed students to designing systems and protocols for large-scale testbeds with several hundred nodes.
