Approaches to grid-based SAT solving

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
Faculty of Information and Natural Sciences | D4 Julkaistu kehittämis- tai tutkimusraportti taikka -selvitys
Degree programme
TKK reports in information and computer science, 16
In this work we develop techniques for using distributed computing resources to efficiently solve instances of the propositional satisfiability problem (SAT). The computing resources considered in this work are assumed to be geographically distributed and connected by a non-dedicated network. Such systems are typically referred to as computational grid environments. The time a modern SAT solver consumes while solving an instance varies according to a random distribution. Unlike many other methods for distributed SAT solving, this work identifies the random distribution as a valuable resource for solving-time reduction. The methods which use randomness in the run times of a search algorithm, such as the ones discussed in this work, are examples of multi-search. The main contribution of this work is in developing and analyzing the multi-search approach in SAT solving and showing its efficiency with several experiments. For the purpose of the analysis, the work introduces a grid simulation model which captures several of the properties of a grid environment which are not observed in more traditional parallel computing systems. The work develops two algorithmic frameworks for multi-search in SAT. The first, SDSAT, is based on using properties of the distribution of the solving time so that the expected time required to solve an instance is reduced. Based on the analysis of SDSAT, the work proposes an algorithm for efficiently using large number of computing resources simultaneously to solve collections of SAT instances. The analysis of SDSAT also motivates the second algorithmic framework, CL-SDSAT. The framework is used to efficiently solve many industrial SAT instances by carefully combining information learned in the distributed SAT solvers. All methods described in the work are directly applicable in a wide range of grid environments and can be used together with virtually unmodified state-of-the-art SAT solvers. The methods are experimentally verified using standard benchmark SAT instances in a production-level grid environment. The experiments show that using the relatively simple methods developed in the work, SAT instances which cannot be solved efficiently in sequential settings can be now solved in a grid environment.
propositional satisfiability, SAT solving, computational grids, distributed search, multi-search
Other note
Permanent link to this item