Embarrassingly Parallel Search in Constraint Programming

Abstract

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

mplrs: A scalable parallel vertex/facet enumeration code

“…Many such problems involve processing an enormous amount of data that can easily be divided, one prominent example being the SETI@home project [1]. A recent approach to parallel constraint solvers [47] (where the input and output are comparatively small) uses this as inspiration and initially creates a large number of subproblems that are then solved in parallel. Other approaches to creating an initial (hopefully balanced) decomposition of the input include cube-and-conquer [39], which uses a lookahead SAT solver to split the original problem into many subproblems that are solved in parallel by CDCL solvers, and applying machine learning techniques to parallel AND/OR branch-and-bound [53].…”

Developing political-ecological theory: The need for Many-Task Computing

“…Therefore, MDAS is not an embarrassingly parallel algorithm. An embarrassingly parallel job (in the sense of “an embarrassment of riches,” see [69]) consists of a set of tasks that can be executed in parallel with no inter-task communication. Also, the objective function evaluation tasks are complex involving for example, the running of a political-ecological simulator many times to support the computation of the consistency analysis objective function.…”

Implementing YewPar: A Framework for Parallel Tree Search

“…There are three main work generation approaches for tree search: 1) Static work generation, as in embarrassingly parallel search [12], creates all tasks at startup and stores them in a (global) workpool, where they are picked up by idle workers. To balance load these approaches need to generate vastly more tasks than the number of workers, which increases startup cost.…”

“…for a specific search application, e.g. Embarrassingly Parallel Search [12] supports constraint programming only; and (2) use hand crafted parallelism, e.g parallel MaxClique [13], with almost no reuse of parallelism between search applications. Hence typically an application is parallelised just once in an heroic effort.…”