ACM logo

The ACM Journal of Experimental Algorithmics


Volume 7, Article 2, 2002


Parallelizing Local Search for CNF Satisfiability Using Vectorization and PVM

by

Kazuo Iwama

and

Daisuke Kawai

and

Shuichi Miyasaki

and

Yasuo Okabe

and

Jun Umemoto

http://www.jea.acm.org/2002/IwamaSatisfiability/

Abstract:

The purpose of this paper is to speed up the local search algorithm for the CNF Satisfiability problem. Our basic strategy is to run some 105 independent search paths simultaneously using PVM on a vector supercomputer VPP800, which consists of 40 vector processors. Using the above parallelization and vectorization together with some improvement of data structure, we obtained 600-times speedup in terms of the number of flips the local search can make per second, compared to the original GSAT by Selman and Kautz. We ran our parallel GSAT for benchmark instances and compared the running time with those of existing SAT programs. We could observe an apparent benefit of parallelization: Especially, we were able to solve two instances that have never been solved before this paper. We also tested parallel local search for the SAT encoding of the class scheduling problem. Again we were able to get almost the best answer in reasonable time.

Received
Accepted
Final Revision
Published




Last updated and validated May 11, 2002, by editor@jea.acm.org