Energy-aware Task Scheduling Strategies for Multi-core Embeddded Systems

Abstract

In this paper, we propose two energy-aware scheduling algorithms---(1) Reinforcement learning-based multiprocessor scheduling (RL) algorithm and (2) Mathematical morphology multiprocessor scheduling (MMS) algorithm---for scheduling time-constrained Directed Acyclic Graph (DAG) tasks in an embedded multiprocessor system with Dynamic Voltage And Frequency Scaling (DVFS) and Dynamic Power Management (DPM) technology. Unlike other heuristic scheduling algorithms, the proposed reinforcement learning (RL) is a machine learning algorithm, rarely considered for energy-aware scheduling in DAG tasks. The MMS, inspired by Mathematical morphology that is often used in image processing, continuously adjusts the coded scheduling through a probe matrix to optimize energy consumption. In this paper the genetic algorithm (GA) is compared with these two proposed algorithms by rigorous simulation. The results demonstrate that our algorithms are more energy efficient. Compared with the GA algorithm, the RL and the MMS algorithm significantly improve the energy consumption reduction rate by an average of 13.37% and 72.92% respectively. In addition, MMS algorithm shows better performance in high-density and high-complexity DAG tasks.