Area-maximizing schedules for series-parallel DAGs

Abstract

Earlier work 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 and massively multicore computers - that 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 and 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. © 2010 Springer-Verlag.