Effectiveness of parallelizing the ILOG-CPLEX mixed integer optimizer in the PUBB2 framework

Abstract

In this paper, we introduce a new method of parallelizing a MIP (Mixed Integer Programming) solver. This method is different from a standard implementation that constructs a parallel branch-and-cut algorithm from scratch (except using an LP solver). The MIP solver we use is ILOG-CPLEX MIP Optimizer (Version 8.0), which is one of the most efficient implementations of branch-and-cut algorithms. The parallelization of the solver is performed by using the software tool PUBB2 developed by the authors. We report a part of our computational experience using up to 24 processors. In addition, we point out some problems that should be resolved for a more efficient parallelization. © Springer-Verlag 2003.