基于物理主机过载检测的虚拟机动态整合方法研究

发布时间：2018-04-03 03:17

本文选题：虚拟机动态整合　切入点：过载检测　出处：《江南大学》2017年硕士论文

【摘要】：随着各种Internet服务不断地移植到公有云平台,云数据中心的规模越来越大,云数据中心作为公有云的数据基础架构面临的挑战也越来越多,不断攀升的能源消耗与云服务质量之间的矛盾就是其中之一。虚拟机迁移和虚拟机动态整合是目前云数据中心虚拟资源管理主要技术和方式。虚拟机迁移是数据中心维持物理主机间负载平衡的主要技术手段,虚拟机动态整合期望能够利用该技术通过迁移虚拟机来优化资源分配,从而达到降低数据中心能耗改善服务质量的目的。在服务质量、资源利用率与能耗开销之间达到一个理想的平衡是云数据中心虚拟资源管理所追求的目标。为此,本文围绕物理主机过载检测、待迁移虚拟机选择和虚拟机放置等与虚拟机动态整合相关的问题进行了研究,主要研究工作和创新一并概括如下:(1)针对设置固定过载阈值的虚拟机动态整合方法难以同时兼顾数据中心能源有效性与服务质量的不足,提出了自适应过载阈值选择AOTS算法。该算法将自适应过载阈值的选择问题建模为马尔可夫决策过程,计算过载阈值的最优选择策略,并根据系统能效和服务质量调整阈值,根据选择的过载阈值检测过载物理主机,并从过载物理主机上选择部分虚拟机重分配至其它物理主机,基于上述研究,提出了基于自适应过载阈值选择的虚拟机动态整合AOTS-VMDC方法,仿真实验表明,同已有方法比较,AOTS-VMDC能够在降低数据中心能耗前提下,提供良好的服务质量。(2)迁移中的虚拟机会停止服务,因此频繁地迁移虚拟机会对服务质量造成影响。对此,基于贝叶斯网络建立了用于估计虚拟机迁移概率的估算模型,用该模型可以预测主机上潜在的虚拟机迁移总次数;由于虚拟机的负载具有随机性,提出利用正态分布来估计物理主机的过载概率,从而检测过载物理主机。基于上述研究,提出了基于贝叶斯网络的虚拟机动态整合BN-VMDC方法,该方法在迁移虚拟机时会优先迁移能够明显降低源主机潜在虚拟机迁移总次数的虚拟机,并将虚拟机优先分配在潜在迁移总次数较小的物理主机上,仿真实验表明,BN-VMDC显著减少了虚拟机的迁移次数,降低了数据中心能耗,并且服务质量获得了明显改善。(3)提出了能耗感知与多资源过载概率约束的虚拟机动态整合EC-VMDC方法。该方法将最小化数据中心能耗、触发的虚拟机迁移次数以及物理主机过载概率作为优化目标。为了搜索到可行的物理主机与虚拟机映射关系,提出一种共享求解经验的随机算法,由于算法能够借鉴已有的求解历史经验来搜索新的可行解,可以用该算法替换人工蜂群的觅食行为,借用蜂群算法的搜索框架,从而降低陷入局部最优的可能性。仿真实验表明,与现有的多资源虚拟机动态整合方法相比,EC-VMDC有效地降低了各项资源过载概率,达到了保证服务质量的目的,此外EC-VMDC在节约电能和减少虚拟机迁移次数方面同样效果显著。
[Abstract]:With a variety of Internet services continue to transplant to the public cloud platform, cloud data center of the increasingly large scale cloud data center as a public cloud data infrastructure challenges are more and more, the contradiction between energy consumption and rising the quality of cloud services is one of them. Migration of virtual machine and virtual machine dynamic integration at present, cloud data center virtual resource management and main technology. Virtual machine migration is the main technical means to maintain the data center physical host load balancing, dynamic virtual machine integration is expected to migrate through the virtual machine to optimize the allocation of resources using the technology, so as to reduce the energy consumption of the data center to improve service quality. The quality of service, achieve an ideal balance between resource utilization and energy consumption is the cloud data center virtual resource management goal. Therefore, this paper The physical host overload detection, the migration of virtual machine and virtual machine placement were studied with virtual machine dynamic integration related issues, the main research work and innovation can be summarized as follows: (1) for the virtual machine dynamic integration method to set fixed overload threshold to both the lack of data center energy efficiency and quality of service the proposed adaptive overload threshold selection AOTS algorithm. This algorithm will be adaptive overload threshold selection problem is modeled as a Markov decision process, optimal selection strategy and according to the calculation of the overload threshold value, the system efficiency and service quality to adjust the threshold according to the overload threshold detection, overload physical host selection, and select the part of the virtual machine weight assigned to the other from the physical host overload physical host, based on the above research, put forward the virtual machine dynamic integration AOTS- adaptive threshold selection based on overload VMDC method, simulation results show that compared with the existing methods, AOTS-VMDC can reduce energy consumption under the premise of the data center, to provide good quality of service. (2) the opportunity to stop the service in virtual migration, so frequent migration of virtual chance to cause influence on the quality of service. In this regard, the Bayesian network is established for estimating model of virtual machine migration based on probability, the model can predict the virtual machine on the host migration potential total number; due to the load of the virtual machine is random, using normal distribution to estimate the probability of overload physical host, so as to detect overload physical host. Based on the above research, put forward a virtual machine dynamic Bayesian network integration method based on BN-VMDC. This method will give priority to migration can significantly reduce the total number of potential migration of virtual machine virtual machine source host in the migration of virtual machine, virtual machine and priority allocation The total number of potential physical host migration smaller, simulation results show that BN-VMDC significantly reduces the number of virtual machine migration, reduces the energy consumption of the data center, and service quality were significantly improved. (3) proposed energy aware and multi resource overload probability constraints of virtual machine dynamic integration method. This method will be EC-VMDC to minimize the energy consumption of the data center, the number of virtual machine migration and physical host overload probability triggering as the optimization target. In order to search the relationship between feasible physical host and virtual machine mapping, this paper proposes a random algorithm for sharing experience, because the algorithm can solve the existing historical experience to search for new solutions, can be replaced with artificial bee colony the algorithm of the foraging behavior of ant colony algorithm use search framework, so as to reduce the possibility of falling into local optimum. Simulation results show that with existing virtual resources Compared with the pseudo dynamic integration method, EC-VMDC effectively reduces the resource overload probability and achieves the purpose of ensuring the quality of service. Besides, EC-VMDC has the same effect in saving energy and reducing the number of virtual machines migration.

【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP302;TP308

【相似文献】