Power Management and Dynamic Voltage Scaling: Myths and Facts
David C. Snowdon, Sergio Ruocco, Gernot Heiser · 2005
This paper investigates the validity of common approaches to power management based on dynamic voltage scaling (DVS). Using instrumented hardware and appropriate operating-system support, we account separately for energy consumed by the processor and the memory system. We find that memory often contributes significantly to overall power consumption, which leads to a much more complex relationship between energy consumption and core voltage and frequency than is frequently assumed. As a consequence, we find that the voltage and frequency setting that minimises energy consumption is dependent on system characteristics, and, more importantly, on the applicationspecific balance of memory and CPU activity. The optimal setting of core voltage and frequency therefore requires either a-priori analysis of the application or, where this is not feasible, power monitoring at run time.