Execution time is no longer the only performance metric for computer systems. In fact, a trend is emerging to trade raw performance for energy savings. Techniques like Dynamic Power Management (DPM, switching to low power state) and Dynamic Voltage and Frequency Scaling (DVFS, throttling processor frequency) help modern systems to reduce their power consumption while adhering to performance requirements. To balance flexibility and design complexity, the concept of Voltage and Frequency Islands (VFIs) was recently introduced for power optimisation. It achieves fine-grained system-level power management, by operating all processors in the same VFI at a common frequency/voltage.This paper presents a novel approach to compute a power management strategy combining DPM and DVFS. In our approach, applications (modelled in full synchronous dataflow, SDF) are mapped on heterogeneous multiprocessor platforms (partitioned in voltage and frequency islands). We compute an energy-optimal schedule, meeting minimal throughput requirements. We demonstrate that the combination of DPM and DVFS provides an energy reduction beyond considering DVFS or DMP separately. Moreover, we show that by clustering processors in VFIs, DPM can be combined with any granularity of DVFS. Our approach uses model checking, by encoding the optimisation problem as a query over priced timed automata. The model-checker Uppaal Cora extracts a cost minimal trace, representing a power minimal schedule. We illustrate our approach with several case studies on commercially available hardware.
展开▼
