Leveraging performance and power savings for embedded systems using multiple target deadlines

Abstract

Tasks running on embedded systems are often associated with deadlines. While it is important to complete tasks before their associated deadlines, performance and energy consumption also play important roles in many usages of embedded systems. To address these issues, we explore the use of Dynamic Voltage and Frequency Scaling (DVFS), a standard feature available on many modern processors for embedded systems. Previous studies often focus on frequency assignment for energy savings and meeting definite task deadlines. In this paper, we present a heuristic algorithm based on convex optimization techniques to compute energy-efficient processor frequencies for soft real-time tasks. Our novel approach provides performance improvements by allowing definitions of multiple target deadlines for each task. We simulate two versions of our algorithm in MATLAB and evaluate their performance and efficiency. The experimental results show that our strategy leverages performance and energy savings, and can be customized to suit practical applications. © 2010 IEEE.