This project continues development of a data management strategy to organize scientific datasets into online (i.e. "in memory") object-oriented numerical databases. The strategy uses hashing maps based on the spatio-temporal coordinates of the dataset itself and addresses a key issue in interactive data analysis - maintaining a retrieval route for the user between the visual representation on the screen and the actual data. This is accomplished through a set of data structures that manage computed data as "computational objects" at the nodes of a mesh organized as a single plane, a volume of planes, or a timeseries. All references to the data are resolved into queries about the maps. The hashing maps can be resident in a workstation while managing a database that is itself distributed across numerous facilities. Current extensions of this research include: (1) design of parallel storage and searching techniques of the distributed data objects; (2) interactive user specification of filter functions (local or non-local) for graphics and diagnostics; and (3) generalization of management utilities to support conversion of any user specified data set to object-oriented database organization. This research seeks to alleviate the continuing user problem of "data glut" in science and engineering. Equally important are its implications for the development of software strategies that can utilize the complex and diverse hardware developments promised for the next decade in massively parallel computing.
