基于负载预测的OpenStack虚拟机智能管理

发布时间：2018-06-20 03:26

本文选题：云计算 + 虚拟机动态配置　；参考：《浙江大学》2015年硕士论文

【摘要】：在云计算应用日益广泛的今天,提高云服务的质量与数据中心的资源利用率,减少云服务商的运营成本与客户的开销具有及其重要的应用价值。论文在国家863计划支持下,针对虚拟机管理这一问题,研究基于负载预测的OpenStack虚拟机智能管理方法。云环境下负载变化较大,带来云服务质量的下降,论文研究了基于短期负载预测的虚拟机动态配置算法。通过提出对虚拟机资源利用率的预测方法,根据预测值结合机器学习的方法计算虚拟机资源包括CPU、RAM、BandWidth的瓶颈因子,找出影响服务质量的瓶颈资源,并使用轻量级的虚拟机运行时动态配置来扩展瓶颈资源,提高服务质量。实验数据表明该算法能大大提高云环境下应用的服务质量,减少服务的响应时间,提高吞吐率。针对云用户请求负载呈现规律性变化的应用场景,论文研究基于长期负载规律的虚拟机管理方法,通过预测用户的请求数,在满足用户资源需求的前提下,动态改变用户的虚拟机分配从而最小化用户的开销。考虑服务器的整合过程,提出了针对物理服务器的数量、虚拟机迁移数、数据中心的电能消耗三个方面多目标优化的算法LBVMM (Load Based Virtual Machine Management)来决策整合的过程,使物理服务器数量、虚拟机的迁移数以及数据中心的电能消耗都维持在一个较低的水平内。算法从云服务商和客户两方面进行了综合考虑,同时优化了云服务商的运营成本,提高服务质量,并且减少了客户的开销。实验数据表明该算法在减少包括物理服务器的数量,虚拟机的迁移数,数据中心的电能消耗,以及用户的开销方面具有显著的效果。结合上述成果,论文在OpenStack平台上实现了虚拟机智能管理系统,系统集成了基于短期负载的虚拟机动态配置与基于长期负载规律的虚拟机管理两个方面,并结合了包括虚拟机批量部署、多层监控等多项创新功能。
[Abstract]:With the increasing application of cloud computing, improving the quality of cloud service and resource utilization of data center, reducing the operating cost of cloud service provider and the expense of customers have important application value. In this paper, an intelligent management method for OpenStack virtual machine based on load prediction is studied, which is supported by the National 863 Program and aiming at the problem of virtual machine management. The dynamic configuration algorithm of virtual machine based on short-term load prediction is studied in this paper. Based on the prediction method of virtual machine resource utilization, the bottleneck factors of virtual machine resources, including CPUUU Ram BandWidth, are calculated according to the prediction value combined with machine learning, and the bottleneck resources that affect the quality of service are found out. Dynamic configuration of lightweight virtual machine runtime is used to expand bottleneck resources and improve QoS. Experimental data show that the proposed algorithm can greatly improve the service quality, reduce the response time and improve the throughput. In view of the application scenario of cloud user request load showing regular change, this paper studies the management method of virtual machine based on long-term load law. By predicting the number of user requests, we can meet the requirements of user resources. Dynamically change the user's virtual machine allocation to minimize the user's overhead. Considering the integration process of servers, this paper proposes a multi-objective optimization algorithm, LBVMM load based Virtual Machine Management, which aims at the number of physical servers, the number of virtual machines and the power consumption of data centers, to make the number of physical servers. The number of virtual machine migrations and the power consumption of the data center are maintained at a low level. The algorithm considers the cloud service provider and customer comprehensively, optimizes the operating cost of the cloud service provider, improves the service quality, and reduces the cost of the customer. The experimental results show that the proposed algorithm can reduce the number of physical servers, the migration of virtual machines, the power consumption of data centers, and the cost of users. Combined with the above achievements, this paper implements a virtual machine intelligent management system on OpenStack platform. The system integrates two aspects: the dynamic configuration of virtual machine based on short-term load and the management of virtual machine based on long-term load rule. And combined with the virtual machine batch deployment, multi-layer monitoring and other innovative functions.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP302

【相似文献】