基于收益的云环境虚拟机资源动态分配方法研究

发布时间：2018-04-13 03:00

本文选题：云计算 + 虚拟化技术　；参考：《东北大学》2012年硕士论文

【摘要】：随着分布式计算、并行计算、和网格计算的发展,云计算开始形成并不断地完善。云计算是基于虚拟化技术,将IT资源构成一个动态的虚拟资源池,以服务的形式供外界使用。虚拟化技术能够有效进行服务整合以减少服务器数量从而降低功耗,增强服务器的可靠性,并能显著简化IT基础设施、优化资源以及降低风险。然而虚拟化技术同时带来的底层物理架构的变化和虚拟机上用户服务的持续变化给云计算环境中底层的物理资源的优化分配方法带来了前所未有的挑战。由于虚拟机上的用户服务的持续变化,虚拟机监视器无法监测到每台虚拟机的各类资源的实际使用量,更无法预测到用户服务的性能状态,如何合理有效地完成资源池中的资源调度分配以提高计算资源的利用率及满足虚拟机上用户服务的性能要求是云计算环境中资源分配研究的重要课题。本文围绕基于虚拟化的云计算进行了广泛的理论与技术研究,通过研究掌握了虚拟化的云计算技术,阐述了云环境下虚拟机资源分配方法的分类和性能预测模型的分类,详细论述了基于收益的云环境虚拟机资源分配的过程,针对虚拟机上的用户服务的性能预测和基于收益的虚拟机资源分配两个问题展开论述和研究,主要提出了基于随机梯度回归的服务性能预测算法和基于收益的云环境虚拟机资源动态分配方法。主要内容如下。首先,考虑云端资源提供者的收益和虚拟机上的用户服务的性能问题,提出了基于收益的云环境虚拟机资源动态分配过程,给出基于收益的云环境虚拟机资源动态分配框架；其次,针对虚拟机上的用户服务的性能预测,利用数据预处理方法对服务性能指标数据进行预处理,采用基于随机梯度回归的服务性能预测模型预测服务的平均响应时间变化情况,为是否进行虚拟机资源分配提供依据；再次,针对虚拟机上的用户服务所需要的各类资源是不确定的,本文采用灰色预测的方法对虚拟机上的用户服务对各类资源的预需求量进行预测,为资源分配提供数据依据；最后,根据用户的预算和平均响应时间的要求,给各个虚拟机上的用户服务划分服务级别,基于收益最大化的原则,提出基于收益的云环境虚拟机资源动态分配方法,根据云端资源提供者满足每个虚拟机上的用户服务的服务质量要求,得到不同的收益(增加或减少),建立资源与云端资源提供者的收益的关系,动态的分配虚拟机资源,提高资源的利用率。本文针对上述内容中的用户服务的性能预测算法-基于随机梯度回归的服务性能预测算法进行了仿真实验,进行了对比实验,结果显示该算法的具有较高的精度,可以为虚拟机的资源分配提供可靠的依据。针对本文提出的基于收益的云环境虚拟机资源动态分配方法,通过对基础云计算平台Hadoop中的虚拟机资源动态分配的需求实例的研究,对虚拟机对各类资源的预需求量的预测验证实例、边际收益和边际成本求解过程验证实例、各个运行用户服务的虚拟机的各类资源的分配验证实例进行定量的验证和分析,得出的结果是本文提出的基于收益的云环境虚拟机资源动态分配方法是可行的和有效的。
[Abstract]:With the development of distributed computing, parallel computing and grid computing, cloud computing began to form and constantly improve. Cloud computing is based on virtualization technology, IT resources to form a virtual resource pool is a dynamic, in the form of services for external use. The virtualization technology can effectively reduce the number of servers and service integration to reduce power, enhance the reliability of server, and it can significantly simplify IT infrastructure, optimize resources and reduce risk. However, changes of virtualization technology and the underlying physical architecture and virtual machine users continued to change the underlying computational physics resources environment to the cloud distribution optimization method has brought hitherto unknown challenge due to the continued. The change on the virtual machine service, virtual machine monitor cannot monitor the actual usage of various resources of each virtual machine, but can not forecast to The performance status of user service, how to efficiently and efficiently allocate and allocate resources in resource pool to improve the utilization of computing resources and satisfy the performance requirements of user services on virtual machines is an important topic in resource allocation research in cloud computing.
The virtual cloud computing based on the theoretical and technical study in this dissertation, through research and master virtual cloud computing technology, expounds the classification of virtual machine resource allocation method under cloud environment classification and performance prediction model, discussed in detail the process of cloud environment virtual machine resource allocation based on income, according to the performance the prediction on the virtual machine user service and two problems of virtual machine resource revenue allocation based on discussion and research, mainly put forward the service performance of stochastic gradient regression prediction algorithm and dynamic resource allocation method based on cloud virtual machine environment based on revenue. The main contents are as follows.
First, considering the revenue of cloud resource providers and the performance of user services on virtual machines, we propose a revenue based dynamic allocation process of virtual machine resources in cloud environment, and give a revenue based dynamic allocation framework of virtual machine resources in cloud environment.
Secondly, according to the performance of virtual machine service users on the service performance index prediction, using data preprocessing method of data preprocessing, the average response time of change of service performance of stochastic gradient regression models for forecasting service based on the situation, whether virtual machine resource allocation;
Thirdly, all kinds of resources needed for user service on virtual machine are uncertain. In this paper, we use grey prediction method to predict the pre demand of user service and various resources on virtual machine, so as to provide data basis for resource allocation.
Finally, according to the user's budget and the average response time to the user service requirements, service level division of each virtual machine, based on the principle of profit maximization, the dynamic resource allocation method of cloud virtual machine environment based on revenue, according to the cloud resource providers meet each virtual machine users on the service quality of service requirements, get different income (increase or decrease), the relationship between the establishment of resources and cloud resource provider revenue, dynamic allocation of virtual machine resources, improve the utilization rate of resources.
Algorithm of service performance prediction algorithm based on stochastic gradient regression simulation experiment was carried out to predict the performance of the content of user services, conducted a comparative experiment results show that the algorithm has high accuracy, provide a reliable basis for the allocation of resources to the virtual machine. The dynamic resource allocation method in the cloud environment virtual machine based on revenue is put forward in this paper, through the research of virtual machine computing of dynamic resource allocation in Hadoop based on cloud platform needs examples, examples of the pre forecast demand for virtual machines of all types of resources, examples the marginal revenue and marginal cost calculation, examples and allocate resources of virtual machines running each user service the verification and quantitative analysis, the result is the proposed dynamic resource allocation method of cloud virtual machine environment based on revenue is Good and effective.

【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP302

【参考文献】