xEnergy-Optimal Control of Multi-Core Applications

Software applications for mobile platforms require a delicate balance between throughput and resource consumption. Programming them is a complex multi-level job: Tasks must be assigned to heterogeneous processors, and must be scheduled on time, to meet throughput constraints. The processors must be controlled for minimal power consumption, using techniques like DPM (Dynamic Power Management) and DVFS (Dynamic Voltage and Frequency Scaling). How to guarantee that the system meets all throughput constraints, while minimizing the energy consumption? This paper reviews our previous work, which proposed to model streaming applications by Synchronous Data Flow, and hardware characteristics by a Processor Platform Model. Both models are transformed into a network of Timed Automata. Subsequently, we use the tool-suite of UPPAAL for analysis and synthesis of schedules with guaranteed throughput and optimal energy consumption. We will discuss several extensions to this basic framework, using Priced Timed Automata and Stochastic Hybrid Automata.