Memory Efficient Parallel SAT Solving with Inprocessing

Abstract

Automatic heuristic configuration and algorithm selection can tremendously improve performance in industrial use-cases of SAT solving. In contrast to attempting to select the best heuristic for the problem, portfolio approaches in parallel SAT solving run different heuristics and even algorithms in parallel. This kind of diversification can be very successful because different heuristics and heuristic configurations have better runtimes on different problems. However, such approaches often suffer from high memory consumption. We present a parallel portfolio SAT solver that is based on several totally different branching heuristics and configurations. In contrast to similar approaches, our portfolio solver uses a shared clause database. We show how to asynchronously manage concurrent access to a shared clause database in a parallel portfolio of solvers that can also perform inprocessing.