A Risk-Constrained Markov Decision Process Approach to Scheduling Mixed-Criticality Job Sets
Résumé
We consider the problem of scheduling Mixed Criticality (MC) job systems with an arbitrary number of criticality levels on a single processing platform, when job demands are probabilistic and their distributions are known. We develop a probabilistic framework for MC scheduling, where feasibility is defined as the risk of missing deadlines, which we express in terms of (chance) constraints on the probabilities that jobs of every criticality miss their deadlines. Our goal is to identify and compute " efficiently implementable " scheduling policies under which the given probabilistic constraints are satisfied. We model the problem as a Constrained Markov Decision Process (CMDP), and we show that a feasible Markov randomized scheduling policy exists if the given instance is feasible in a probabilistic sense. A feasible policy can be obtained by solving a linear program. To counter the potential state space explosion, we outline an approximation method that might trade feasibility for efficiency, but which performs well in practice.
Domaines
Informatique [cs]
Origine : Fichiers produits par l'(les) auteur(s)
Loading...