Scalable data handling in sensor networks

Sensor networks are an important class of distributed systems which combine distributed sensing, computation, storage, and wireless multi-hop communication. Numerous scientific and commercial applications that have emerged in recent years, and a large number of industrial and research institutions working in this area.; This thesis focuses on overcoming the storage constraints of small form-factor wireless platforms in emerging sensor networks. Many currently available prototype sensor nodes such as MICA Motes [URLa] have a storage capacity of a few Megabytes and must constrain communication in order to conserve energy and extend lifetime.; This thesis explores systems that provide a lossy, gracefully degrading storage model. We believe that such a model is necessary and sufficient for many scientific applications since it supports both progressive data collection for interesting events as well as long-term in-network storage for in-network querying and processing.; A gracefully degrading storage model can provide multiple benefits for future long-term sensor network deployments for high-bandwidth applications. By storing sensor data in a hierarchical multi-resolution manner within the network, it provides an efficient search mechanism for user-queries that process past data. In addition, it provides a framework for in-network processing techniques such as identification of long-term trend or anomalous features in data. We expect that older data can be stored with sufficient fidelity within the network to satisfy such long-term queries at significantly lower cost than centralized data collection. Our implementation of long-term storage and in-network data aging has been done on the Emstar development framework at UCLA based on Linux-XScale platforms.; A progressive data collection approach benefits scientists who would like an interactive system to identify and store new kinds of event signatures for future analysis. An instance of such an application is a structural vibration data acquisition system that focuses on real-time data gathering of structural vibration data from short-term deployments such as shaker tables. Multi-hop data gathering from these systems incurs large latency, for instance, collecting 15 minutes (200KB at each node) of vibration data from a network of 20 motes typically involves latencies of four to eight hours. In a progressive data collection model, low-resolution summaries of the event are transmitted to the base station with low-latency. These summaries can be analyzed by scientists and, if required, high-resolution lossless event data can be gathered before it is aged out from the local store on nodes. An implementation of progressive-lossy data gathering is available for the Mica2 motes as part of the structural data acquisition project at the Center for Embedded Networked Sensing (CENS).