强化学习在虚拟机资源调度中的应用

Application of Reinforcement Learning in Virtual Machine Resource Scheduling

非统一内存访问(NUMA,Non-Uniform Memory Access)体系结构因其可扩展性而被广泛应用于虚拟化和云计算中。在NUMA系统中,以前的工作主要关注如何通过调度来减少远端内存访问以优化系统性能。然而,共享资源的竞争也是影响虚拟化NUMA系统性能的重要因素之一。针对NUMA架构下由于虚拟机放置而产生的共享资源竞争问题,建立了初始虚拟机放置模型,针对该模型提出了一种基于强化学习的算法Post来求解。实验结果表明,该算法能够有效地降低执行时间,在准确率上优于传统的基于策略优化算法,可以达到提升系统性能的目的。

The non-uniform memory access(NUMA)architecture is widely used in virtualization and cloud computing due to its scalability.In the NUMA system,previous work mainly focused on how to reduce remote memory access through scheduling to optimize system performance.However,competition for shared resources is also one of the important factors affecting the performance of virtualized NUMA systems.Aiming at the shared resource competition problem caused by virtual machine placement under the NUMA architecture,we established an initial virtual machine placement model.A reinforcement learning-based algorithm Post is proposed to solve the model.Experimental results show that the algorithm can effectively reduce the optimization time,better than traditional algorithm based on strategy optimization in accuracy,and can achieve the goal of improving system performance.