Abstract. We propose the Embarrassingly Parallel Search, a simple and efficient method for solving constraint programming problems in parallel. We split the initial problem into a huge number of independent subproblems and solve them with available workers (i.e., cores of machines). The decomposition into subproblems is computed by selecting a subset of variables and by enumerating the combinations of values of these variables that are not detected inconsistent by the propagation mechanism of a CP Solver. The experiments on satisfaction problems and on optimization problems suggest that generating between thirty and one hundred subproblems per worker leads to a good scalability. We show that our method is quite competitive with the work stealing approach and able to solve some classical problems at the maximum capacity of the multi-core machines. Thanks to it, a user can parallelize the resolution of its problem without modifying the solver or writing any parallel source code and can easily replay the resolution of a problem.

International audienceWe introduce an Embarrassingly Parallel Search (EPS) method for solving constraint problems in parallel, and we show that this method matches or even outperforms state-of-the-art algorithms on a number of problems using various computing infrastructures. EPS is a simple method in which a master decomposes the problem into many disjoint subprob-lems which are then solved independently by workers. Our approach has three advantages: it is an efficient method; it involves almost no communication or synchronization between workers; and its implementation is made easy because the master and the workers rely on an underlying constraint solver, but does not require to modify it. This paper describes the method, and its applications to various constraint problems (satisfaction, enumeration, optimization). We show that our method can be adapted to different underlying solvers (Gecode, Choco2, OR-tools) on different computing infrastructures (multi-core, data centers , cloud computing). The experiments cover unsatisfiable, enumeration and optimization problems, but do not cover first solution search because it makes the results hard to analyze. The same variability can be observed for optimization problems, but at a lesser extent because the optimality proof is required. EPS offers good average performance, and matches or outperforms other available parallel implementations of Gecode as well as some solvers portfolios. Moreover, we perform an in-depth analysis of the various factors that make this approach efficient as well as the anomalies that can occur. Last, we show that the decomposition is a key component for efficiency and load balancing

International audienceWe propose an adaptation of the Embarrassingly Parallel Search (EPS) method for data centers. EPS is a simple but efficient method for parallel solving of CSPs. EPS decomposes the problem in many distinct subproblems which are then solved independently by workers. EPS performed well on multi-cores machines (40), but some issues arise when using more cores in a datacenter. Here, we identify the decomposition as the cause of the degradation and propose a parallel decomposition to address this issue. Thanks to it, EPS gives almost linear speedup and outperforms work stealing by orders of magnitude using the Gecode solver

International audienceThis paper presents an external parallelization of Constraint Programming (CP) search tree mixing both static and dynamic partitioning. The principle of the parallelization is to partition the CP search tree into a set of sub-trees, then assign each sub-tree to one computing core in order to perform a local search using a sequential CP solver. In this context, static partitioning consists of decomposing the CP variables domains in order to split the CP search tree into a set of disjoint sub-trees to assign them to the cores. This strategy performs well without adding an extra cost to the parallel search, but the problem is the load imbalance between computing cores. On the other hand, dynamic partitioning is based on preservation of the search state to generate, dynamically or on demand, the sub-trees that are assigned to the cores. This strategy offers good load balancing between the different computing cores, but computing overcosts appear due to the initialisation of the search when a sub-tree is migrated from one core to another. In this paper, we propose a new partitioning strategy that mixes the static and dynamic partitioning and enjoys the benefits of each strategy. This mixed partitioning is designed to run on shared and distributed memory architectures. The performances obtained are illustrated by solving the CP problems modelled using the FlatZinc format and solved using the Google OR-Tools solver on top of the parallel Bobpp framework