Optimising earliest deadline first scheduling for parallel real-time tasks on multiprocessors

Abstract

Multiprocessors have become prevalent in real-time systems owing to their higher throughput. Various types of scheduling algorithms have been proposed for parallel real-time tasks, which differ from traditional tasks in that their subtasks execute in parallel. A parallel task is frequently modelled as a directed acyclic graph (DAG) that expresses the precedence constraints between its subtasks. In this Letter, the authors propose a decomposition algorithm to improve the Earliest Deadline First schedulability for DAG tasks, based on convex optimisation theory. Their experimental results demonstrate that their algorithm outperforms the two most recently published algorithms.