The federated scheduling of systems of conditional sporadic DAG tasks

Abstract

A federated approach to the multiprocessor scheduling of systems of independent recurrent tasks is considered, in which each task is either restricted to execute preemptively upon a single processor, or may execute upon multiple processors but gets exclusive access to all these processors. Efficient polynomial-time algorithms are derived here for the federated schedulability analysis and run-time scheduling of recurrent task systems that are represented by the conditional sporadic DAG tasks model. The performance of these algorithms is characterized via a speedup factor metric, which quanti es the combined cost of both restricting oneself to the federated scheduling paradigm, and of requiring the scheduling algorithms to run in polynomial time.