Parallel Heuristic Search

Tito Autrey and Herbert Gelernter

SYNCHEM is a large expert system in the domain of organic chemistry. It finds synthetic pathways by chaining backwards from the target molecule to available compounds. SYNCHEM uses heuristic search to explore the solution space efficiently. Depending on the complexity of the target compound, the system currently takes from a few hours to several solve an interesting problem using the present knowledge base. A production sized knowledge base would increase the search time by an order or magnitude or more. This is an unacceptable response time if SYNCHEM is to become a practical tool for organic chemists. Many organic chemists now have access to powerful workstations for tasks such as molecular modeling, information retrieval and data analysis. Some of these are shared memory multi-processor (SMP) machines. When groups of workstations are networked together they can be viewed as a distributed memory (DM) machine. Since the inter-process communication costs are different for SMP and DM machines, they require different approaches to harnessing all of their computational power. Users of SYNCHEM are interested in speedy results and in quality results. It is necessary to enhance the search strategy algorithms to take into account the architecture-dependent communication costs. These algorithms should be scalable to large numbers of processors, and should be able to make dynamic tradeoff between quality of search and time to solution.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.