Power management in the EPOS system

Abstract

Power management strategies for embedded systems typically rely on static, application driven deactivation of components (e.g. sleep, suspend), or on dynamic voltage and frequency scaling. However, the design and implementation of these strategies in embedded operating system often fail to deal with real-time and quality-of-service (QoS) requirements. The EPOS system implements an infra-structure that supports both static (application-driven) and dynamic (system-driven) power management. In this work, this infrastructure is used to explore energy as a parameter for QoS in embedded systems, with the goal of guaranteeing energy consumption metrics, while preserving the deadlines of essential (hard real-time) tasks. Given a set of real-time tasks and their associated energy consumption, we provide equations to check schedulability in project-time. At runtime, a preemptive scheduler for imprecise tasks prevents the execution of optional subtasks whenever there is the possibility of deadline loss or depletion of the energy source. We show that this mechanism is effective in controlling energy consumption and ensuring "best-effort" computation without deadline loss.