Principled Tuning And Structuring Methods For Sensornets

Wireless Sensor Networks, or sensornets, are an emerging class of information processing system. Unlike conventional computer networks, sensornets compose many independent motes into self-organising ad hoc networks. Motes are small, cheap computers, each equipped with independent power supplies, wireless communication capability, and sensors with which to passively monitor the physical environment in which they are embedded. Distributed applications distil voluminous raw data about sensed physical phenomena into meaningful information with utility to end users. All nodes are equal peers, responsible for data production, processing, storage, delivery and consumption. Management is decentralised. Interactions with the real world imply real-time requirements. Designing and configuring sensornets is difficult. Motes have limited resources, limited connectivity with peers, and are liable to fail. Unstable physical environments imply variation in sensor data volume and content, and variation in wireless connectivity. Countless configurations of application software, network topologies, middleware, and protocols are possible, with no guarantee that any combination is sufficiently performant and reliable. This thesis aims to contribute toward understanding, and solving, the problems associated with designing large scale self-managing sensornets. We argue that properties of sensornet behaviour can be measured and quantified such that objective evaluation and comparison among the set of candidate configurations is feasible. Significant improvements in measurable attributes of sensornet behaviour can be obtained through appropriate design decisions in network protocol selection and logical configuration. Designing custom protocols for specific deployments offers the potential for improved performance. However, unknown and unforeseeable properties of these new protocols may result in diminished performance or incorrect behaviour, potentially leading to system failure. Instead, existing protocols can be tuned for reuse in specific deployments. This retains the benefits of predictable behaviour, established both in theory and by experience accumulated from previous deployments, while providing improved performance. Existing protocols can be tuned for specific deployments, maximising performance while retaining confidence of correct behaviour accumulated from previous experience. Where protocol tuning cannot deliver the required behaviour as a consequence of inherent properties of protocols, logical networks overlaid on physical networks enable further improvement, providing a platform in which protocols can fulfil application requirements.