Whole-System Optimality Analysis and Tailoring of Worst-Case–Constrained Applications (Watwa)

Energy-constrained real-time systems, such as implantable medical devices, are prevalent in modern life. These systems demand its software to fulfill both properties of safe and energy-efficient task executions. Regarding safety, these systems must execute their tasks within execution-time and energy bounds since resource-budget violations potentially cause danger to life. In order to guarantee the system's safe execution with the given time and energy resources, static program-code analysis tools are automatically able to determine the system's worst-case resource-consumption behavior. However, existing static analyses so far are not able to tackle the problem of resource-efficient execution while maintaining the property of safe execution under the given resource constraints. Achieving the system's efficient execution through manual tailoring would involve an unrealistically high effort, especially when considering the large amount of energy-saving features in modern system-on-chip (SoC) platforms. In order to eventually yield resource-optimal execution and likewise allow the operating system to safely schedule tasks, a whole-system view on software tasks, their resource constraints, and hardware features would be essential, which goes beyond the current state of the art.

The research proposal Watwa describes an approach for whole-system optimality analysis and automatic tailoring of worst-case-constrained applications. The core idea is the automatic generation of variants of the analyzed application that are equivalent from a functional point of view. The variant generation accounts for the multitude of modern energy-saving features, which, in turn, allows subsequent optimizing analyses to tailor the application by aggressively switching off unneeded, power-consuming components in the SoC. The temporal and energetic behavior of these variants is examined by means of worst-case analysis tools that yield bounds on the respective resource demands, which eventually achieves the safe execution during runtime. This novel combination of the variant generation and the analysis of their worst-case behavior allows Watwa to systematically determine hints for safe, resource-optimal scheduling sequences. In order to exploit these hints during runtime, the project proposes an operating system along with its scheduler that dynamically reacts to changes in the environment and exploits the determined scheduling hints for efficiency considerations while enforcing safe operation within resource budgets. In summary, the goal of this project is to provide answers to the following two questions: (1) How can program-code analyses determine resource-optimal task variants by exploiting modern hardware features while accounting for worst-case bounds? (2) How can operating systems exploit analytically determined scheduling hints on possible task-execution sequences and hardware activations to guarantee safe task executions while increasing the system's efficiency?