Adaptive DVS management through elastic scheduling

Dynamic voltage scaling (DVS) is a technique used in modern microprocessors operated by battery to set voltage and frequency levels at proper values that meet performance requirements while minimizing energy consumption. Most of the present work on DVS management, however, is based on simplistic assumptions about the hardware characteristics that limit the real applicability of the proposed algorithms. Typical simplifying assumptions consider continuous processor speed, negligible overhead during voltage switching, task execution time linear with frequency, and tasks with equal power consumption. In this work, we enhance the task model to consider some of the real CPU characteristics, and integrate energy-aware algorithms with elastic scheduling to improve control performance of embedded systems running on architectures offering a limited number of operating modes. Implementation issues and experimental results for the proposed algorithm are also discussed