Energy-efficiency for multiframe real-time tasks on a dynamic voltage scaling processor

Abstract

Energy-aware design for electronic systems has been an important issue in hardware and software implementations. Dynamic voltage scaling (DVS) techniques have been adopted to effectively trade the performance for the energy consumption. However, most existing research for energy-efficient design in DVS systems with realtime constraints focuses on tasks with worst-case execution times. Once a task instance completes earlier than its worst-case estimation, the unused slacks can be used for slowing down to reduce the energy consumption. This paper explores how to efficiently and effectively minimize the energy consumption to schedule a set of periodic real-time tasks with the multiframe property, in which the execution times of task instances are characterized by a vector of elements that are repeated. This paper proposes two types of approaches: (1) the task-based approach and (2) the frame-based approach. The task-based approach allocates the same time length for the executions of task instances belonging to the same task. The frame-based approach can reduce the energy consumption further by assigning an execution speed to each task frame. For on-line use, the scheduling overhead for speed determination is constant in both types of the proposed approaches. Simulations show that our proposed approaches sacrifice some optimality in terms of energy savings, compared to the optimal solutions, but require less space and less overhead for scheduling in the on-line (run-time) fashion.