Donald Bren School of Information and Computer Sciences

As computing moves to battery operated portable systems, the functionality is increasingly implemented in software with an embedded/real-time operating system (RTOS). For such systems, there is a need for power-aware applications and system software. In this paper, we present a layered software architecture that enables the application and OS programmers to design energy-efficient applications and RTOS services. The software architecture consists of a power-aware RTOS kernel and a set of standard software interfaces that enable easy exchange of timing and power information among the underlying hardware platform, the RTOS, and the applications. To demonstrate the utility of our approach we focus on making the task scheduling process in an RTOS power-aware, and incorporate an OS-directed dynamic power management technique that enables adaptive power-fidelity tradeoffs during task scheduling. We have implemented it using the RedHat eCos operating system running on a complete variable voltage system based on the Intel XScale micro-architecture. We ran four different algorithms, from a simple shutdown based scheme to a dynamic predictive and adaptive DVS algorithm. The results show an energy gain of up to 66% when comparing to the execution without any power management incorporated and 17% comparing to the simple shutdown scheme.