SelfSplit parallelization for mixed-integer linear programming

SelfSplit is a simple mechanism to convert a sequential tree-search code into a parallel one. In this paradigm, tree-search is distributed among a set of workers, each of which is able to autonomously determine—without any communication with the other workers—the job parts it has to process. SelfSplit already proved quite effective in parallelizing Constraint Programming solvers. In the present paper we investigate the performance of SelfSplit when applied to a Mixed-Integer Linear Programming (MILP) solver. Both ad-hoc and general purpose MILP codes have been considered. Computational results show that SelfSplit, in spite of its simplicity, can achieve good speedups even in the MILP context. The role of performance variability in limiting scalability is also discussed.