Management and search of large databases is a computational bottleneck in many domains. One attractive approach to eliminating it is to exploit recent advances in parallel computer hardware and processing. However, the development of software that permits this represents one of the major challenges facing computer scientists. This project proposes to develop a generic approach to parallel database management that should be especially suitable for large databases of complex objects in which even single object comparisons may be computationally costly. (Examples include databases of image data, in which database search may require numerous comparisons of complicated images.) This approach is based on the use of Linda, a commercial product of Scientific Computing Associates, Inc. Linda systems provide convenient access to parallelism from within familiar high-level languages such as C or Fortran. In Phase I, a fast text retrieval system will be developed for asynchronous multiprocessors. This will serve as proof of concept for more general software to be investigated in Phase II work. Due to the use of Linda, the technology developed will be not only efficient, but also portable across vastly different parallel architectures including both explicitly parallel computers and local-area networks of workstations functioning as parallel "hypercomputers".
