Rules class approach to scheduling algorithms

Abstract

As processors on-chip gain in numbers and complexity, task scheduling has become an important concern in system design, and the related research has produced substantial and diversified knowledge. As a result, the efficient taping and management of this knowledge has become a concern in itself. In particular, it can bring new ways to improve scheduling algorithms. This paper describes a new algorithm class based on association rules mining. It serves to both increase the knowledge about a particular scheduling algorithm and show how to improve its performance. Two examples show how this new methodology can be used to improve makespan and processor use globally by optimizing the scheduling method locally.