Mixed Criticality Systems with Weakly-Hard Constraints

Mixed criticality systems contain components of at least two criticality levels which execute on a common hardware platform in order to more efficiently utilise re- sources. Due to multiple worst-case execution time estimates, current adaptive mixed criticality scheduling policies assume the notion of a low criticality mode where by a taskset executes under a set of more realistic temporal assumptions and a high criticality mode, in which all low criticality tasks in the taskset are descheduled, to ensure that high criticality tasks can meet more conservative timing constraints derived from certification approved methods. This issue is known as the service abrupt problem and comprises the topic of this work.
The principles of real-time schedulability analysis are first reviewed, providing relevant background and theory on which mixed criticality systems analysis is based. The current state-of-the-art of mixed criticality systems scheduling policies on uni-processor systems are then discussed along with the major challenges facing the adoption of such approaches in practice. To address the service abrupt issue this work presents a new policy, Adaptive Mixed Criticality - Weakly Hard which provides a guaranteed minimum quality of service for low criticality tasks in the event of a criticality mode change. Two offline response time based schedulability tests are derived for this model and dominance relationship proved. Empirical evaluations are then used to assess the relative performance against previously published policies and their schedulability tests, where the new policy is shown to offer a scalable performance trade-off between existing fixed priority preemptive and adaptive mixed criticality policies. The work concludes with possible directions for future research.