Dynamic voltage and frequency scaling based parallel scheduling scheme for video recognition on multicore systems

Abstract

The video recognition on computer systems has become a high computing complexity and consumed the enormous amount of energy. This paper uses DVFS power management technique schedules the frame task to cores with higher or lower VF settings depending on a task's time-varying compute intensity. The DVFS based scheduling scheme minimize the number of voltage and frequency transition to solve the limitation of traditional DVFS for multicore systems. The proposed scheme can further improve the energy consumption by 9% to 16% with great speedup.