Flexibility and Portability for Embarrassingly Parallel Search

EasyChair Preprint no. 2815

Abstract

Constraint programming is a paradigm for solving combinatorial problems. Yet solving complex problems may be a lengthy process. Embarrassingly Parallel Search (EPS) is "a simple and efficient method for solving constraint programming problems in parallel". While EPS is a very generic method, its implementations are strongly dependent on the machine architecture for which they were written. We suggest a new communication architecture for EPS's inner workings such that it becomes usable on different machine architectures. Then we exhibit a prototype that combines partitioning and portfolio methods implementing that architecture.

Keyphrases: Parallélisme, Programmation par contraintes, Solveur de contraintes

@Booklet{EasyChair:2815,
  author = {Samvel Balassanian Dersarkissian and Arnaud Malapert},
  title = {Flexibility and Portability for Embarrassingly Parallel Search},
  howpublished = {EasyChair Preprint no. 2815},

  year = {EasyChair, 2020}}