SatX10: A Scalable Plug&Play Parallel SAT Framework

Learning Sequential and Parallel Runtime Distributions for Randomized Algorithms

“…In particular, we use the feature computation code available from the Beta lab at the University of British Columbia. 1 This tool can compute several collections of features describing the SAT instances. Some are static, such as: number of variables, clauses, ratio variables/clauses, etc.…”

“…, D 10 }. For each subset D i∈ [1,10] , the machine learning model is learned with D −D i and tested on D i . The advantage of the 10-fold cross-validation is that all instances in the same problem family are used for both training and testing, but each instance is used for testing exactly once.…”

“…Now, with the improvements of the solvers and the availability of computation time in parallel architectures, that can be rented by the hour, the problem has changed. The question, as in [1] for instance, is no longer "how much time would I need to solve an instance", but rather "since I have paid for x hours of computation on n cores, how many SAT instances could I solve?". In this paper, we propose a method for solving large classes of SAT instances with randomized algorithms, by learning features related to the solving time on a small set of instances and applying this knowledge to predict the solving time for the other instances.…”

Parallel Combinatorial Optimization with Decision Diagrams

“…It supports parallelism natively and applications built with it can be compiled to run on various operating systems and communication hardware. Similar to SatX10 [11], we capitalize on the fact that X10 can incorporate existing libraries written in C++ or Java. We start off with the sequential version of the BDD code base for MISP [7] and integrate it in X10, using the C++ backend.…”

“…For example, Bloom et al [11] recently introduced SatX10 as an efficient and generic framework for parallel SAT solving using X10. We refer to our proposed framework for parallel decision diagrams as DDX10.…”