Distributed Operating Systems on Wireless Sensor Networks

This thesis proposes the use of traditional distributed operating system and distributed systems techniques that are adapted and applied to the wireless sensor network domain. These techniques are applied to the creation of a wireless sensor network operating system that allows complex applications to be created without special programmer knowledge of sensor network programming or architecture. The resulting system is capable of executing a high level user application written in conventional single-system-image form, without the user being aware of the mesh architecture or underlying sensor node hardware. A wireless sensor network is a collection of battery-powered embedded systems that communicate over low-bandwidth radio. Because of their limited hardware, niche deployments and use of embedded processors, programming techniques for wireless sensor network nodes are generally relatively esoteric compared to most software programming tasks. This can be relatively complex for programmers not familiar with the wireless sensor network domain. A naive approach to writing a wireless sensor network application may well result in considerably reduced battery life due to inefficient use of the limited power resources, requiring an expensive and time-consuming replacement or patching process. As a result of this complexity, traditional wireless sensor network applications are written as simply as possible. The majority of these applications simply move passive data readings back across a mesh to a more powerful server. While this is a sufficiently effective approach in some situations, for other sensor network deployments involving large amounts of complex data it is more efficient for the sensor network application to process at least some of the data inside the mesh, saving on unnecessary data transmissions. However in the real world, the complexity of writing such an application in many cases precludes this from being created. An operating system that provides power-efficient distributed processing while presenting a more standard unified single system image to the application developer would provide new possibilities for sensor network application developers in terms of creating dynamic and complex sensor network applications. This thesis covers the design decisions, development process and evaluation of the Hydra distributed wireless sensor network operating system, an operating system that provides these services. The system is evaluated in the form of a scenario for monitoring intruders over a large area using accelerometer monitoring -- during this scenario, power efficiency is gained due to the intelligent Hydra operating system services, as the resulting accelerometer data is not moved across potentially multi-hop network links. Application code complexity is also reduced due to the higher-level single system image programming environment