REEG: Reinforcement Learning-Based Gang Scheduling in Multicore Embedded Systems

In modern cyber-physical systems, the increasing complexity and parallel execution demands of applications necessitate adopting advanced task scheduling and task orchestration models to optimize system performance and resource allocation in multicore systems. In this brief, we present a novel task model and its corresponding RL-based algorithm, named Reinforcement Learning-Based Energy-Efficient Gang Scheduling in Multicore Cyber-Physical Systems (REEG) in which each node of a Directed Acyclic Graph (DAG) is characterized as a gang task, requiring simultaneous execution across multiple cores, thus offering a precise and scalable framework for optimizing the performance of complex, highly parallel workloads in modern multicore systems. However, significant energy optimization challenges exist due to the complex dependencies and simultaneous core usage. We propose a framework based on reinforcement learning (RL) that dynamically modifies core allocation and execution techniques to reduce energy consumption and maintain computational efficiency without compromising quality of service (QoS) or performance. This approach reduces energy consumption while maintaining performance and QoS, ensuring efficient computation. Our experimental results demonstrate that the RL-based method achieves notable energy reductions and improves system efficiency compared to the state-of-the-art method. On average, our proposed method (REEG) achieves 28.19% less energy consumption and 60.75% more QoS compared to the state-of-the-art methods.