Empirical Study of the Anatomy of Modern Sat Solvers

Abstract

Boolean Satisfiability (SAT) solving has dramatically evolved in the past decade and a half. The outcome, today, is manifested in dozens of high performance and relatively scalable SAT solvers. The significant success of SAT solving technology, specially on practical problem instances, is credited to the aggregation of different SAT enhancements. In this paper, we revisit the organization of modern conflict-driven clause learning (CDCL) solvers, focusing on the principal techniques that have contributed to their impressive performance. We also examine the interaction between input instances and SAT algorithms to better understand the factors that contribute to the difficulty of SAT benchmarks. At the end, the paper empirically evaluates different SAT techniques on a comprehensive suite of benchmarks taken from a range of representative applications. The diversity of our benchmarks enables us to make fair conclusions on the relation between SAT algorithms and SAT instances.