基于博弈论的虚拟资源竞拍分配研究

发布时间：2018-10-18 16:22

【摘要】：虚拟化技术的出现改变了管理计算机操作系统、存储以及应用程序的方式，它提供了共享物理资源的能力，使得物理资源能够在同一时间内被多个不同的虚拟机使用，消除了操作系统对物理硬件的依赖性。随着云计算的兴起，云环境下的资源管理成为人们研究的热点，而虚拟化技术作为云计算中的关键技术日益显示出其资源管理方面的独特优势。然而，在利用虚拟化技术进行资源管理时也遇到了一些挑战，其中的一个挑战就是多虚拟机之间资源分配的公平性。有效合理的虚拟机资源分配是虚拟化及云计算环境下系统性能及服务质量保证的关键。在虚拟化环境中当多个虚拟机共存于同一物理机时，由于物理资源的有限性，可能会发生激烈的资源竞争。传统虚拟机资源分配通常没有考虑这种竞争给系统造成的影响，而是按照物理资源数量进行均值分配或者系统对资源的峰值需求来进行分配。这些分配方式没有考虑到各虚拟机自身的负载情况，往往会造成资源的浪费。本文针对这一问题进行了研究，运用经济学中博弈论的思想提出了基于惩罚机制的CPU竞拍模型和基于市场机制的内存预测分配模型。本文针对Xen虚拟化平台主要做了如下几个方面的工作：本文首先系统介绍了虚拟机以及虚拟化的相关技术，详细阐述了虚拟化技术的分类，并且简要介绍了每种虚拟化技术的主要产品以及这些产品的优缺点，而且还对这五种虚拟化技术进行了比较。然后重点分析了半虚拟化产品Xen的资源管理的特点，并在此基础上将博弈论的思想引入到Xen虚拟资源分配中，分别提出了CPU和内存的博弈分配方案。在Xen中，CPU资源一旦分配给某一虚拟机即可投入使用，针对这一特征，本文提出了基于激励惩罚机制的博弈竞拍模型。在这个模型中，本文选取应用程序的响应时间和CPU利用率作为系统的评价指标，并且采用博弈竞拍的思想探讨了CPU分配与竞争对这些指标的影响关系，然后利用合作博弈的NASH均衡进行求解多虚拟机的动态资源分配问题。为验证这种方案的有效性，我们通过实验对模型的稳定性和公平性进行了验证，通过对比发现采用博弈控制之后比单纯的平均分配资源更能实现按需分配资源，提高资源利用率，这样才能真正体现资源的公平分配。在Xen虚拟平台下，内存资源通过Ballooning机制分配给虚拟机，有着比较严重的延迟效应。针对这个问题，本文提出了基于市场机制的内存预测分配模型。该模型中，首先我们采用基于马尔科夫链残差GM(1,1)灰色预测模型来预测未来一段时间虚拟机的内存使用量，然后虚拟机管理器根据这些预测值，按照虚拟机拥有的股份按比例地分配与回收内存资源。最后通过系统仿真发现这一模型能够提高物理内存在虚拟操作系统中的利用率，实现内存的按需分配。本文提出的虚拟资源分配方案是面向大规模服务器集群的，具有一般性，可以推广到网络带宽、I/O已经其他计算资源方面的分配与管理工作。
[Abstract]:the advent of virtualization technologies has changed the way in which computer operating systems, storage, and applications are managed, providing the ability to share physical resources so that physical resources can be used by multiple different virtual machines at the same time, the dependency of the operating system on the physical hardware is eliminated. With the rise of cloud computing, resource management in cloud environment becomes the hot spot of people's research, while virtualization technology as the key technology in cloud computing is showing its unique advantages in resource management. However, some challenges have been encountered in managing resource management with virtualization technologies, one of which is the fairness of resource allocation between multiple virtual machines. Effective and reasonable resource allocation is the key to system performance and service quality assurance in virtualized and cloud computing environments. When multiple virtual machines coexist in the same physical machine in a virtualized environment, intense resources may occur due to the physical resources of the physical resource Competition. Traditional virtual machine resource allocation usually does not take into account the impact of this competition on the system, but rather the mean distribution according to the physical resource quantity or the peak demand of the system. Assignment. These allocations do not take into account the load situation of each virtual machine itself, which often results in resources In this paper, the paper studies the problem and put forward the CPU model based on penalty mechanism and the allocation of memory prediction based on market mechanism by using the idea of game theory in economics. Model. This paper focuses on the following aspects of Xen virtualization platform: Work: This article first introduces the virtual machine and the related technologies of virtualization, and explains the virtualization in detail The technical classification, and briefly introduces the main products of each virtualization technology as well as the advantages and disadvantages of these products, and also provides the five virtualization technologies Then, the characteristics of the resource management of the semi-virtualized product Xen are analyzed, and then the idea of game theory is introduced into the Xen virtual resource allocation, and the CPU and the memory are respectively put forward. In Xen, once the CPU resources are allocated to a certain virtual machine, it can be put into use. In view of this feature, this paper proposes an incentive-based penalty machine. In this model, the response time and CPU utilization rate of the application program are chosen as the evaluation index of the system, and the CPU allocation and competition are discussed with the idea of game theory. Using NASH equilibrium of cooperative game to solve multi-virtual machine In order to verify the validity of this scheme, we validate the stability and fairness of the model by experiment. Through the comparison, we find that it is more effective to realize the allocation of resources according to need than the simple sharing resources after game control. and improve the utilization rate of the resources so that the real body can be obtained. under the Xen virtual platform, the memory resources are distributed to the virtual machine through the Ballooning mechanism, There is a serious delay effect. In view of this problem, this paper puts forward a market based on the market In this model, firstly we use the Markov chain residual GM (1,1) gray prediction model to predict the memory usage of the virtual machine in the future, and then the virtual machine manager according to these predicted values, according to the proportion of the shares owned by the virtual machine. Finally, the system simulation shows that this model can improve the utilization of virtual operating system in physics. The virtual resource allocation scheme proposed in this paper is a large-scale server cluster, it has generality, can be extended to network bandwidth, I/ O is already othe
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【参考文献】