Intelligent Task Scheduling for Multi-Core Processors Based on Graph Neural Networks and Deep Reinforcement Learning

Within heterogeneous multi-core systems, the parallelization of tasks and efficient utilization of processors are of paramount significance. The interdependent tasks are often represented as directed acyclic graphs (DAGs), and an effective task scheduling algorithm can significantly optimize the performance of these systems. However, the complex interactions between processors with different processing capabilities and interdependencies between tasks make task scheduling an NP-complete challenge. As artificial intelligence developed, Deep Reinforcement Learning (DRL) provides an effective solution to this challenge. In this paper, we propose an intelligent DRL-based task scheduling model for heterogeneous multi-core systems, named GG-DRL, aiming at minimizing the processing time makespan and maintaining load balancing between processors. Firstly, we adopt Markov Decision Process to model the scheduling problem. Secondly, we design a graph preprocessing module combining Graph Attention Network (GAT) and Gated Recurrent Unit (GRU) to deal with the complex DAG task graph structure to enhance the representation of task features. Finally, we use an Actor-Critic reinforcement learning strategy to train this model effectively. Experiments demonstrate that the model can achieve a shorter makespan as well as a better load balancing effect.