Performance Evaluation of Q-Learning-Based List Scheduling Algorithms in Heterogeneous Systems

Task scheduling plays a crucial role in Heterogeneous Systems (HS). The makespan (or scheduling length) minimization is the key criterion of the task scheduling algorithm in HS. Recently, the rapid development of machine learning has led to its widespread application in various fields. This paper extends existing heuristic-based task scheduling algorithms using Reinforcement learning techniques. Moreover, we use the Q-learning technique to extend two heuristic scheduling algorithms, the PEFT and IPPTS. We aim to combine these algorithms with Q-learning to improve their task prioritization. We compare the HEFT, QL-HEFT, QL-PEFT, QL-PPTS, and QL-IPPTS using real-world scientific workflows. The experimental results show that QL-HEFT outperforms QL-PEFT, QL-PPTS, and QL-IPPTS by 27.12 percent, 16.75 percent, and 10.7 percent, respectively