SRLB: 一种基于两阶段强化学习的负载均衡任务调度算法

he issue of task scheduling in cloud environments is a well-explored challenge. In load balancing task scheduling, currently available reinforcement learning methods usually use a weighting approach to balance load balancing metrics and time metrics, and the pre-specified weights are difficult to adapt to the variable cloud environment. To address the limitations, we formulate a constrained optimization problem on the basis of policy learning methods and propose a two-stage reinforce learning method, TSRLB, as a solution to the load balancing task scheduling problem. Through a trace-driven simulation, we demonstrate that the TSRLB algorithm outperforms traditional algorithms and existing reinforcement learning methods in two metrics: mean completion time and load balancing.