Experience of parallel AI programming with parallel Lisp》.pdfVIP

Future Generation Computer Systems7 (1991/92) 211-219 211 North-Holland Experience of parallel AI programming with parallel Lisp Hiroshi G. Okuno Basic ResearchLaboratories, Nippon Telegraphand TelephoneCorporation, 3-9-11Midoricho, Musashino, Tokyo 180Japan Abstract Okuno, H.G., Experience of parallel AI programmingwith parallel Lisp, Future Generation Computer Systems7 (1991/92) 211-219. Parallelism or parallel execution is expected to improvethe performance of Artificial Intelligence (AI) systemsso that they can be applied to much wider areas. One of the major problems with parallelizing AI systemsis the lack of methodologyfor parallel AI programming. This paper discusses key issues in parallel AI programmingwith parallel Lisp. By implementing two AI systems,OPS5(a rule-based system)and ATMS (an intelligent database system),three main problems are observed: difficulty in identifyingthe most time-consumingsmall tasks, frequent accessto global data, and over-sequential execution. Solutions to these problems are presented, including hierarchical decompositionof tasks, runtime control of multiprocess- ing, reader-writer locks and lazy and speculative computations. Keywords. Parallel Lisp; OPS5; ATMS; run-time control; dynamic spawning; parallel AI; reader-writer lock; hierarchical decomposition. 1o Illtr.duction significantly sped up if they are to be applied to new increasingly complex and time-critical do- Parallelism or parallel execution is expected to mains. We focus our attention on a specific pro- improve the performance of