Opportunistic scheduling in a constraint-rich world

Abstract

The latest planners and schedulers allow expressive domain modelling and problem definition, particularly with respect to the inclusion of constrained resource usage and inter-task dependencies. This increased complexity removes the ability to guarantee schedulability of a problem at run-time.In hard real-time systems, where 'hard' emphasizes the critical nature of meeting task deadines, the estimated worst-case execution time is used in the task representation. If a solely static framework is used to schedule these systems, the pessimism in the prediction of task execution times will lead to unused resources.This paper describes a framework to include a local dynamic scheduler with a large-scale planner or scheduler. The local scheduler can take advantage of any unused resources by scheduling additional task sets. It can also handle online plan repair by switching between different quality levels of tasks.