Bambi Meets Godzilla: Object Databases for Scientific Computing

摘要

Object-oreiented databases (OODBs) are in many ways a better match for scientific data management than conventional record-oriented database systems. User-defined databases reduce the encoding going from a scientific domain to the database. Direct support for complex objects is useful for capturing hierarchical structures, such as molecules. OODBs generally have collection types, such as lists and arrays, that are a better basis than sets for the dimensional data common in scientific applications. Their inherent extensibility seem a good match for handling new kinds of metadata, and having behavior definable in the database permits transparent access to existing data in multiple formats via a common object model.We begin by recounting out experiences with using OODBs for scientific data, in the domains of computational chemistry, and matericals science. The bulk of the talk, however, deals with areas that nee improvements for OODBs to support scientific applications well, among them: Management of massive data sets and tertiary storage Data loading and archiving Querying over ordered collection toys Availability on appropriate computng platforms Application programming interfaces, particularly for FORTRAN and parallel environments Supporting data interchange formats