Towards Class-Based Dynamic Voltage Scaling for Multimedia Applications

Video display has significant, but highly variable, CPU requirements. As such, it is an attractive target for power management via dynamic voltage scaling. In previous work, we have proposed a dynamic voltage scaling algorithm directed to the context of video kiosks, in which a minimal frequency for each frame can be determined experimentally based on observations taken during the first few iterations of the video. In this paper, we review that work, and begin to consider how such an approach can be adapted to the more common case where a video is only played once, on hardware that is not known in advance