A New Perspective on Criticality: Efficient State Abstraction and Run-Time Monitoring of Mixed-Criticality Real-Time Control Systems (Artifact)

The increasing complexity of real-time control systems, comprising control tasks interacting with physics and non-control tasks, comes with substantial challenges: meeting various non-functional requirements implies conflicting design goals and a pronounced gap between worst and average-case resource requirements up to the overall timeliness being unverifiable. Mixed-criticality systems (MCS) are a well-known mitigation concept that operate the system in different criticality levels with timing guarantees given only to the subset of critical tasks. In many real-world applications, the criticality of control applications is tied to the system’s physical state and control deviation, with safety specifications becoming a crucial design objective. Monitoring the physical state and adapting scheduling is inaccessible to MCS but has been dedicated mainly to control engineering approaches such as self-triggered (model-predictive) control. These, however, are hard to schedule or expensive at run time. This paper explores the potential of linking both worlds and elevating the physical state to a criticality criterion. We, therefore, propose a dedicated state estimation that can be leveraged as a run-time monitor for criticality mode changes. For this purpose, we develop a highly efficient one-dimensional state abstraction to be computed within the operating system’s scheduling. Furthermore, we show how to limit abstraction pessimism by feeding back state measurements robustly. The paper focuses on the control fundamentals and outlines how to leverage this new tool in adaptive scheduling. Our experimental results substantiate the efficiency and applicability of our approach.