Decntr: Optimizing Safety and Schedulability with Multi-Mode Control and Resource Allocation Co-Design

As cyber-physical systems (CPS) become increasingly autonomous, there is a growing need for resource-efficient design techniques that can guarantee safety and timeliness during system reconfiguration or mode changes. In this paper, we present Decntr, a co-design technique for jointly optimizing safety, schedulability and robustness for multi-mode CPS on multi-core platforms. By designing switching controllers that can switch be-tween different implementations and between different sampling periods, Decntr gives the resource allocation significantly more flexibility to adapt scheduling decisions to load changes, such as additional tasks in a new mode or increased demands during a mode change. For example, it can pick the best implementation for a task depending on the current resource availability; it can adapt the period within a safe range to effectively utilize CPU and shared resources, which helps increase performance and robustness; and it can relax some job deadlines to avoid transient overloads during mode transitions for better schedulability. Our evaluation on an automotive case study and resource-intensive benchmarks shows that Decntr is highly effective in maximizing schedulability and robustness while ensuring safety, and that it significantly outperforms the state of the art in multi-core resource allocation for multi-mode systems.