Distributed and Self-Organizing Data Management Strategies for Wireless Sensor Networks. A Cross-Layered Approach.

Over the past few decades the computing industry has gone past several milestones, each of which has had a paradigm shift in the way we live our lives. Computers were initially only confined to large corporations. With the advent of personal computers, people started using them daily at work and also at home. Today it is not uncommon for a person to move around with several devices which have substantial amounts of computation power, e.g. a notebook, mobile phone, PDA, digital camera, navigation system, etc. While one may be inclined to feel that we are already surrounded by a huge number of embedded devices, it appears that the computing industry, thanks to the further miniaturization of electronics, might be on the verge of another paradigm shift - one that would make computers omnipresent. The past decade has seen the emergence of a new breed of tiny computers known as wireless sensor nodes. These nodes, which may be battery powered, are equipped with sensors, a radio transceiver, a CPU and some memory. They are usually networked together to form wireless sensor networks. It is envisioned that sensor networks made up of hundreds, thousands or probably even millions of nodes will eventually weave into the very fabric of our lives and be present in one form or another in even the most mundane of devices, like in a coffee mug for instance. The enormous scale of these networks makes it impossible for them to be managed manually by humans. In other words, the system needs to operate autonomously and recover automatically from faults that may occur. As sensor nodes are typically highly energy constrained devices, network lifetime is also of paramount importance. Unlike conventional computer networks, e.g. an office LAN, which can be used for a multitude of applications, wireless sensor networks are generally known to be application-specific. This unique characteristic helps save energy as it allows protocols designed for sensor networks to be optimized for a particular application. This thesis focuses on different techniques that may be used to extract data from a wireless sensor network in an energy-efficient manner. We present a range of distributed, self-organizing and energy-efficient data management algorithms that influence different components of the sensor node architecture: MAC, routing, data aggregation and sensor sampling. The algorithms we present are used for two different classes of applications: (i) applications that only require a subset of all the data in the network to be extracted using range-queries and (ii) applications that require all the data to be extracted from all the sensors in the network at periodic intervals using long-running queries. For the first class of applications, we present a framework that ensures that one-shot range queries are routed only to the relevant regions of the network instead of carrying out flooding. The same framework is also used to assign an appropriate amount of bandwidth to regions that are expected to generate more data with respect to the incoming query. This allows precious energyresources to be spent only where gains are expected. For the second class of applications, we have designed two algorithms that help extract raw data in an energy-efficient manner. The first algorithm, that takes advantage of spatial correlations that may exist between the readings of neighbouring sensor nodes, is a scheduling algorithm which decides when a particular node should aggregate data. The second algorithm helps save energy by sampling the sensors in an energy-efficient manner by taking advantage of temporal correlations that may exist between successive sensor readings. In every instance, we have illustrated how the various algorithms can benefit by using cross-layer information.