Parallellogic Programming and Parallel System Software and Hardware

Abstract

Abstract : This constitutes the final report of work performed under AFOSR grant number 90-0027 to investigate parallel problem solving and deductive databases. Under the grant experiments were performed on the PRISM parallel inference system on the BBN Butterfly. The experiments evaluated alternative message passing strategies for distributing tasks to processors at run-time. Several enhancements were made to PRISM during the grant period. These are: a new inference engine was implemented which provides more efficient support for the full control language of PRISM; and a stack based inference engine was implemented which provides efficient support for the use of limited set of control strategies. Simulation studies were performed which evaluate alternative methods for scheduling tasks on parallel architectures. Two methods were examined which allow the OR-parallel execution of logic programs with no communication overhead. A study was performed evaluating two alternative methods for incorporating integrity constraints into query processing in PRISM. In the first method, separate constraint processors are introduced which check constraints at run-time. In the second method, constraints are incorporated through compile-time transformations. The study indicates that constraints are useful inquiry processing and that the compile-time methodology results in more efficient performance than checking constraints at run-time. In addition to the above, work continued in the area of informative answers to queries in deductive databases.