Dramatic increases in computing power and storage capacity have allowed scientists and engineers to model nature in more detail than ever. However, such models often require massive input and output datasets, and the size of these datasets is outpacing our ability to manipulate and use them. In response, the project is developing a novel approach (Computational Database Systems) for exploring and understanding massive scientific datasets. The basic idea is to integrate scientific datasets stored as spatial databases with specialized, tightly-coupled, and highly-optimized functions that query and manipulate these databases. In particular, the project is developing computational database systems for building and querying the massive datasets produced by unstructured finite element simulations. The project expects to make the following major contributions that span scientific computing, database systems, and storage systems: (1) New algorithms for balancing linear octrees based on new techniques known as "balance by parts" and "prioritized ripple propagation". (2) Efficient new sorting-based techniques for extracting mesh structure from linear octrees. (3) Minimizing output data volume through compression of "near zero" simulation outputs. (4) New techniques for indexing compressed spatio-temporal datasets. (5) Storage management techniques that automatically determine the best layout for data at each level of the system's memory hierarchy.
