Date of Award

Fall 2004

Project Type


Program or Major

Computer Science

Degree Name

Doctor of Philosophy

First Advisor

R Daniel Bergeron


The principal goal of this research was to develop a formal comprehensive model for representing highly complex scientific data. An effective model should provide a conceptually uniform way to represent data and it should serve as a framework for the implementation of an efficient and easy-to-use software environment that implements the model. The dissertation work presented here describes such a model and its contributions to the field of scientific databases. In particular, the Granite model encompasses a wide variety of datatypes used across many disciplines of science and engineering today. It is unique in that it defines dataset geometry and topology as separate conceptual components of a scientific dataset. We provide a novel classification of geometries and topologies that has important practical implications for a scientific database implementation. The Granite model also offers integrated support for multiresolution and adaptive resolution data. Many of these ideas have been addressed by others, but no one has tried to bring them all together in a single comprehensive model.

The datasource portion of the Granite model offers several further contributions. In addition to providing a convenient conceptual view of rectilinear data, it also supports multisource data. Data can be taken from various sources and combined into a unified view.

The rod storage model is an abstraction for file storage that has proven an effective platform upon which to develop efficient access to storage. Our spatial prefetching technique is built upon the rod storage model, and demonstrates very significant improvement in access to scientific datasets, and also allows machines to access data that is far too large to fit in main memory. These improvements bring the extremely large datasets now being generated in many scientific fields into the realm of tractability for the ordinary researcher.

We validated the feasibility and viability of the model by implementing a significant portion of it in the Granite system. Extensive performance evaluations of the implementation indicate that the features of the model can be provided in a user-friendly manner with an efficiency that is competitive with more ad hoc systems and more specialized application specific solutions.