A hierarchical group model for programming sensor networks

A hierarchical group model that decouples computation from hardware can characterize and aid in the construction of sensor network software with minimal overhead. Future sensor network applications will move beyond static, homogeneous deployments to include dynamic, heterogeneous elements. These sensor networks will also gain new users, including casual users who will expect intuitive interfaces to interact with sensor networks. To address these challenges, a new computational model and a system implementing the model are presented. This model ensures that computations can be readily (re)assigned as sensor nodes are introduced or removed. The model includes methods for communication to accommodate these dynamic elements. This dissertation presents a detailed description and design of a computational model that resolves these challenges using a hierarchical group mechanism. In this model, computation is tasked to logical groups and split into collective and local components that communicate hierarchically. Local computation is primarily used for data production and publishes data to the collective computation. Similarly, collective computation is primarily used for data aggregation and pushes results back to the local computation. Finally, the model includes data-processing functions interposed between local and collective functions and are responsible for data conversion. This dissertation also presents implementations and applications of the model. Implementations include Kensho, a C-based implementation of the hierarchical group model, that can be used for a variety of user applications. Another implementation, Tables, presents a spreadsheet-inspired view of the sensor network that takes advantage of hierarchical groups for both computation and communication. Users are able to specify both local and collective functions that execute on the sensor network via the spreadsheet interface. Applications of the model are also explored. One application, FUSN, provides a set of methods for constructing filesystem-based interfaces for sensor networks. This demonstrates the general applicability of the model as applied to sensor network programming and management interfaces. Finally, the model is applied to a novel privacy algorithm to demonstrate that the model isn't strictly limited to programming interfaces.