Area-Maximizing Schedules for Series-Parallel DAGs

Earlier work has introduced a new optimization goal for DAG schedules: the “AREA” of the schedule. AREA-maximizing schedules are intended for computational environments—such as Internet-based computing (IC for short) and massively multicore computers—that would benefit from DAG-schedules that produce execution-eligible tasks as fast as possible. The earlier study of AREA-maximizing schedules showed how to craft such schedules efficiently for DAGs that have the structure of trees, as well as other, less well-known, families of DAGs. The current paper extends the earlier work by showing how to efficiently craft AREA-maximizing schedules for series-parallel DAGs, a family that arises, e.g., in multi-threaded computations. The tools that produce the schedules for series-parallel DAGs promise to apply also to other large families of computationally significant DAGs.