Modern Cooperative Parallel SAT Solving

Abstract

Nowadays, powerful parallel SAT solvers are based on an algorithm portfolio. The alternative approach, (iterative) search space partitioning, cannot keep up, although, according to the literature, iterative partitioning systems should scale better than portfolio solvers. This rises often! In this paper we identify key problems in current parallel cooperative SAT solving approaches, most importantly communication, how to partition the search space, and how to utilize the sequential search engine. First, we improve on each problem separately. In a further step, we show that combining all the improvements leads to a state-of-the-art parallel SAT solver, which does not use the portfolio approach, but instead relies on iterative partitioning. The experimental evaluation of this system completely changes the picture about the performance of search space partitioning SAT solvers: on instances of a combined benchmark of recent SAT competitions, the presented approach can keep up with the winners of last years SAT competition. The combined improvements improve the existing cooperative solver splitter by 24%: instead of 561 out of 880 instances, the new solver Pcasso can solve 696 instances.