Adaptive DAG Tasks Scheduling with Deep Reinforcement Learning

• Published in: Algorithms and Architectures for Parallel Processing Vol. 11335; pp. 477 - 490
• Main Authors: Wu, Qing, Wu, Zhiwei, Zhuang, Yuehui, Cheng, Yuxia
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2018; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: DAG scheduling; Deep reinforcement learning; Heterogeneous
• ISBN: 9783030050535; 303005053X
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-05054-2_37

Efficient task scheduling is critical for improving system performance in the distributed heterogeneous computing environment. The DAG (Directed Acyclic Graph) tasks scheduling problem is NP-complete and it is hard to find an optimal schedule. Due to its key importance, the DAG tasks scheduling problem has been extensively studied in the literature. Many previously proposed heuristic algorithms are usually based on greedy methods, which still exists large optimization space to be explored. In this paper, we proposed an adaptive DAG tasks scheduling (ADTS) algorithm using deep reinforcement learning. The scheduling problem is properly defined with the reinforcement learning process. Efficient scheduling state space, action space and reward function are designed to train the policy gradient-based REINFORCE agent. Leveraging the algorithm’s capability of exploring long term reward, the ADTS algorithm could achieve good scheduling policies. Experimental results showed the effectiveness of the proposed ADTS algorithm compared with the classic HEFT/CPOP algorithms.