Reducing Loss of Service for Mixed-Criticality Systems through Cache-and Stress-Aware Scheduling

Hardware resources found in modern processor architecture, such as the memory hierarchy, can improve the performance of a task by anticipating its needs based on its execution history and behaviour. Interleaved jobs, belonging to other tasks with different behaviours, can cause stress on those resources by disrupting the execution history thus slowing down more sensitive tasks. Schedulability analyses and policies tend to ignore such behaviours, in favour of conservative assumptions, as their effects are difficult to assess. When they are included, the analysis can be very complex and the measures needed are hard to obtain. In this paper, we propose abstract timing models that capture stress and sensitivity with respect to the memory hierarchy. The advantage of an abstract timing model is that it can be derived through measurements without the need for a detailed understanding of the precise cache hierarchy and how it affects software execution. The disadvantage of course is that there is no hard timing guarantee especially if timing anomalies may exist. The contribution of this paper is to build on existing priority assignment schemes using the timing model to discriminate between tasks within sharing a priority levels and improve the system’s timing behaviours in overload. More specifically, we show that for task sets scheduled with mixed-criticality scheduling the number of jobs not executed is often reduced.