摘要
随着大数据时代的来临,工作流应用开始由原有的基础设施转移到更加高效、可靠、廉价的云平台上。本文针对现有基于强化学习的云工作流调度算法收敛速度慢的问题,详细分析云工作流任务的执行流程,设计了一种细粒度的云工作流系统模型,提出了一种结合自适应自然梯度高斯过程回归和强化学习的云工作流调度算法。算法采用强化学习的任务分配算法解决虚拟机间负载均衡问题,并通过自适应自然梯度高斯过程回归加速最优策略的生成。在Workflow Sim平台下进行了验证,实验结果证明了本文算法在一定程度上加速了最优策略的收敛。
With the arrival of big data age,workflow applications are transferring from original infrastructure to more efficient,reliable and affordable cloud computing platforms.Aim at the problem of slow convergence of reinforcement learning based cloud workflow schedule algorithm,we detail analyzed the execute process of cloud workflow jobs,and designs a fine cloud workflow system,then proposed a cloud workflow scheduling algorithm based on adaptive nature gradient Gaussian Process Regression and Reinforcement Learning.We apply the reinforcement-learning based scheduling algorithm to balance the Virtual Machine loads,ant the optimal scheme can be accelerate obtaining by using adaptive nature gradient Gaussian process regression.We tested our algorithm by using WorkflowSim and the experimental results demonstrated the scheme can accelerate the convergence to a certain extent.
作者
钟积海
崔得龙
ZHONG Ji-hai;CUI De-long(College of Computer and Electronic Information,Guangdong University of Petrochemical Technology Maoming 525000,China;Key Project of Guangdong Province in the Research Center of Cloud Robot(Petrochemical)Engineering Technology,Maoming 525000,China)
出处
《电子设计工程》
2018年第16期35-39,44,共6页
Electronic Design Engineering
基金
国家自然科学基金项目(61672174)
广东省云机器人(石油化工)工程技术研究中心开放基金(201606A02)
茂名市石油化工自动化工程技术研究开发中心开放基金
广东石油化工学院大学生创新创业培育计划项目(2016py A032)
关键词
云计算
云工作流
强化学习
高斯过程回归
Q值函数近似
cloud computing
cloud workflow
reinforcement learning
gaussian process regression
Qvalue function approximation