Reducing the solution space of optimal task scheduling

Many scheduling problems are tackled by heuristics due to their NP-hard nature. Task scheduling with communication delays (P|prec,c"i"j|C"m"a"x) is no exception. Nevertheless, it can be of significant advantage to have optimal schedules, e.g. for time critical systems or as a baseline to evaluate heuristics. A promising approach to optimal task scheduling with communication delays for small problems is the use of exhaustive search techniques such as A^@?. A^@? is a best first search algorithm guided by a cost function. While good cost functions reduce the search space, early results have shown that problem specific pruning techniques are paramount. This paper proposes two novel pruning techniques that significantly reduce the search space for P|prec,c"i"j|C"m"a"x. The pruning techniques Fixed Task Order and Equivalent Schedules are carefully investigated based on observations made with simple graph structures such as fork, join and fork-join, yet they are generally applicable. An extensive experimental evaluation of computing more than two thousand schedules demonstrates the efficiency of the novel pruning techniques in significantly reducing the solution space.