Predictive models for multimedia applications power consumption based on use-case and OS level analysis

Power management at any abstraction level is a key issue for many mobile multimedia and embedded applications. In this paper a design workflow to generate system-level power models will be presented, tailored to support quantitative run-time power optimization policies to be implemented within an operating system. The approach we followed to derive power models is strongly use-case oriented. Starting from a comprehensive general and accurate model of a representative architecture for embedded applications (including a multi core MPSoC, accelerators, interfaces and peripherals), a methodology to derive compact models is presented, based upon the distinctive characteristics of the selected use cases. The methodology to generate such model, whose exploitation is foreseen within a power manager working at the OS level, is the focus of the paper. The value and accuracy of the approach is quantitatively and statistically justified through extensive experiments carried out on a development board designed for multimedia applications.