Semi-partitioned mixed-criticality scheduling

Paper presented at the 30TH INTERNATIONAL CONFERENCE ON ARCHITECTURE OF COMPUTING SYSTEMS (ARCS 2017). 3 to 6, Apr, 2017, Session 6: Scheduling. Vienna, Austria.Scheduling isolation in mixed-criticality systems is challenging without sacrificing performance. In response, we propose a scheduling approach that combines server-based semi-partitioning and deadline scaling. Semi-partitioning (whereby only some tasks migrate, in a carefully managed manner), hitherto used in single criticality systems, offers good performance with low overheads. Deadline-scaling selectively prioritizes high-criticality tasks in parts of the schedule to ensure their deadlines are met even in rares case of execution time overrun. Our new algorithm NPS-F-MC brings semi-partitioning to mixed-criticality scheduling and uses Ekberg and Yi’s state-of-the-art deadline scaling approach. It ensures scheduling isolation among different-criticality tasks and only allows low-criticality task migration. We also explore variants that disallow migration entirely or relax the isolation between different criticalities (SP-EKB) in order to evaluate the performance tradeoffs associated with more flexible or rigid safety and isolation requirements.