Storage-Centric Wireless Sensor Networks for Smart Buildings

Abstract

In the first part of the dissertation, we propose a model-based systems design framework, called WSNDesign, to facilitate the design and implementation of wireless sensor networks for Smart Buildings. We apply model-based systems engineering principles to enhance model reusability and collaboration among multiple engineering domains. Specifically, we describe a hierarchy of model libraries to model various behaviors and structures of sensor networks in the context of Smart Buildings, and introduce a system design flow to compose both continuous-time and event-triggered modules to develop applications with support for performance evaluation. WSNDesign can obtain early feedback and high-confidence evaluation of a design without requiring any intrusive and costly deployment. In addition, we develop a graphical tool that exposes a sequence of design choices to system designers, and provides instant feedback about the influence of a design decision on the complexity of system analysis. Our tool can facilitate comprehensive analysis and bring competitive advantage to the systems design workflow by reducing costly unanticipated behaviors.

One of the main challenges to design efficient sensor networks is to collect and process the data generated by various sensor motes in Smart Buildings efficiently. To make this task easier, we provide an abstraction for data collection and retrieval in the second part of the dissertation. Specifically, we design and implement a distributed database system, called HybridDB, for application development. HybridDB enables sensors to store large-scale datasets in situ on local NAND flash using a novel resource-aware data storage system, and can process typical queries in sensor networks extremely efficiently. In addition, HybridDB supports incremental $\epsilon$-approximate querying that enables clients to retrieve a just-sufficient set of sensor data by issuing refinement and zoom-in sub-queries to search events and analyze sensor data efficiently. HybridDB can always return an approximate dataset with guaranteed maximum absolute ($L_\infty$-norm) error bound, after applying temporal approximate locally on each sensor, and spatial approximate in the neighborhood on the proxy. Furthermore, HybridDB exploits an adaptive error distribution mechanism between temporal approximate and spatial approximate for trade-offs of energy consumption between sensors and the proxy, and response times between the current sub-query and the following sub-queries. The implementation of HybridDB in TinyOS 2.1 is transformed and imported to WSNDesign as a part of the model libraries.