Scheduling for Mixed-criticality Hypervisor Systems in the Automotive Domain

This thesis focuses on scheduling for hypervisor systems in the automotive domain. Current practices are primarily implementation-agnostic or are limited by lack of visibility during the execution of partitions. The tasks executed within the partitions are classified as event-triggered or time-triggered. A scheduling model is developed using a pair of a deferrable server and a periodic server per partition to provide low latency for event-triggered tasks and maximising utilisation. The developed approach enforces temporal isolation between partitions and ensures that time-triggered tasks do not suffer from starvation. The scheduling model was extended to support three criticality levels with two degraded modes. The first degraded mode provides the partitions with additional capacity by trading-off low latency of event-driven tasks with lower overheads and utilisation. Both models were evaluated by forming a case study using real ECU application code. A second case study was formed inspired from the Olympus Attitude and Orbital Control System (AOCS) to further evaluate the proposed mixed-criticality model. To conclude, the contributions of this thesis are addressed with respect to the research hypothesis and possible avenues for future work are identified.