【摘要】 关于计算机系统与网络中的资源管理问题的研究无处不在，其中计算集群的调度算法一直是研究的热点。目前大多数解决方案为启发式调度算法，但启发式算法无法全面地感知系统中调度作业之间的关联性，而深度强化学习可以通过数据自主学习这些潜在的关联性。本文使用了一种基于动作分支架构改进的深度强化学习调度算法，在Spark调度模型中取得了不错的效果。该算法通过将一个完整的调度过程分解为相对独立的分支动作，从而简化各个动作设计过程并有效降低动作空间的维度。实验结果表明，在相同的训练时间内，该模型取得了较好的调度性能。更多还原

【Abstract】 Research on resource management in computer systems and networks is ubiquitous, among which the scheduling algorithm of computing clusters has always been a research hotspot. Most of the current solutions are heuristic scheduling algorithms. However, heuristic algorithms cannot fully perceive the correlations between scheduled jobs in the system, while deep reinforcement learning can learn these potential correlations autonomously through data. In this paper, an improved deep reinforcement learning scheduling algorithm based on the action branch architecture is used, and good results have been achieved in the Spark scheduling model. By decomposing a complete scheduling process into relatively independent branch actions, the algorithm simplifies the design process of each action network and effectively reduces the dimension of the action space.Experimental results show that the model achieves better scheduling performance within the same training time.更多还原