Macroprogramming scalable sensor networks
Abstract
Recent prototypical systems have demonstrated applications of wireless sensor networks in diverse sensing and data processing domains. As applications of sensor networks mature, there is an increasing realization of the complexity associated with programming large numbers of resource and energy constrained heterogeneous devices that operate cooperatively in highly dynamic environments. In this thesis, we investigate sensor network macroprogramming: a novel high-level programming model that entails specification of aggregate distributed system behavior. Our macroprogramming architecture facilitates development of self-organized sensor network systems that are scalable, energy-efficient, and verifiable. This model is in contrast to the state of the art that involves programming of low-level behavior of individual nodes, which is a non-trivial, effort-intensive, and error-prone task. We undertake a comprehensive study to address fundamental challenges in macro-programming sensor networks. Specifically, we make the following contributions towards realization of a comprehensive macroprogramming infrastructure. First, we present a high-level programming model and associated infrastructure, COSMOS, which allows explicit specification of distributed coordinated behavior in contrast to the traditional model of programming individual nodes. We present a lean operating system that supports COSMOS applications, and features support for scalable distributed data flow based programs. Second, we examine the complexity of developing applications that adaptively manage resource tradeoffs under constraints, and argue that such applications require static behavioral verification. To this end we propose a high-level logic language and an associated synthesis engine that verifies user programs and generates robust applications. Third, we present a compiler for sensor network applications that generates energy-efficient and low messaging cost primitives from specifications of complex distributed sensornet protocols. Finally, we present an in-network data compression protocol that enables application independent high-resolution sensor monitoring while minimizing the energy footprint. We support claims of efficiency and effectiveness of our macroprogramming infrastructure with experiences from a real-world deployment of our system and comprehensive empirical evaluation.
Degree
Ph.D.
Advisors
Grama, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.