Parallelizing Constraint Solvers for Hard RCPSP Instances

The Resource-Constrained Project Scheduling Problem (RCPSP) is a well-known scheduling problem aimed at minimizing the makespan of a project subject to temporal and resource constraints. Constraint Programming allows to model and solve RCPSPs in a natural and efficient way, especially when Lazy Clause Generation (LCG) techniques are employed. In this paper we show that hard RCPSPs can be efficiently tackled by a portfolio approach that combines the strengths of different —not necessarily LCG-based— solvers by exploiting their parallel execution on multiple cores. Our approach seeks to predict and run the best solvers for a new, unseen RCPSP instance and also enables the bounds communication between them. This on-average allows to outperform the oracle solver that always chooses the best available solver for any given instance