【作者】 李贝贝；

【导师】 贺红；

【作者基本信息】 山东大学 ， 计算机科学与技术， 2016， 硕士

【摘要】 虚拟化技术有效解决了传统数据中心的资源浪费、管理困难等问题,通过建立虚拟机共享物理设备资源,利用有限的物理资源运行多个独立操作系统实现不同功能及任务的分配。虚拟机在多台物理主机之间的迁移技术已经有广泛深入的研究,可以达到负载均衡、热点解除等目标。但是在实际应用中,很多企业的虚拟化数据中心采用静态管理方式,在创建虚拟机时分配固定的资源,运行过程中无法根据负载需求进行动态调控,会造成资源不足的现象；并且随着集群规模的增大,人工管理的效率大大降低。因此,建立集群的实时监控系统,自动监测各节点和虚拟机的运行状态,根据实际需求进行合理的资源分配管理,可以保证数据中心运行时的服务质量,节约人力资源。本文主要研究了虚拟化集群的监控管理系统的设计与实现的关键技术,在总结前人研究工作的基础上,本文所做的主要工作如下：1.结合节点内部的虚拟机资源动态分配技术和节点间的虚拟机迁移技术,设计了集群监控管理系统的框架模型,将系统功能分为状态监控和资源管理模块,集群结构分为中心节点和工作节点,分别负责节点内部监控和集群整体监控,并且当需求增加导致出现热点问题时进行自主管理。2.工作节点实时监控节点内部每个虚拟机的运行状态,随着虚拟机的应用负载增加,其创建时初始化分配的资源不足以满足需求,达到资源使用率阈值时进行报警。文中建立了根据历史记录预测虚拟机负载的模型,根据模型算法确定如何对资源进行动态分配。保证了虚拟机有足够的资源运行应用并且控制资源的分配策略,防止由于资源不足而造成服务质量下降。3.中心节点负责对集群整体进行监控,物理节点上的虚拟机负载增加会造成该节点上的剩余资源不断减少,影响物理机器的运行,无法保证节点上的虚拟机良好的服务质量。中心节点发现热点节点将进行报警和迁移策略的制定。建立了基于迁移开销预测的模型进行虚拟机选择,根据虚拟机的内存和页面修改速度选择待迁移的虚拟机；设计了负载状态感知的目标节点选择算法,根据节点的负载和资源状态确定要迁入虚拟机的目的主机。通过Cloudsim软件设计了合理的仿真对比实验证明了算法的良好效果。更多还原

【Abstract】 Virtualization technology effectively solves the problems such as waste of resources and management difficulties in traditional data center. It implements distribution of different functions and tasks by creating virtual machines to share physical equipment resources and using limited physical resources to run multiple operating systems. The technology of virtual machines’ migration between hosts already has extensive and further research which can achieve the goal of load balancing and hot release. But in application, many businesses’virtualized data center adopts static management of recourses that allocates quota resources when creating virtual machines. It cannot adjust dynamically according to the load demands at runtime which would cause lack of resources. Besides, with the increasing scale of clusters, the efficiency of artificial management will greatly reduce. Accordingly, it can improve service quality of the active data center and economize human resources to build a real-time monitoring system of cluster, automatically monitor the running state of all nodes and virtual machines, and make a reasonable resource allocating management based on real demand.In this paper, we studied the key technologies in designing and implementation of the monitoring and management system of virtualized cluster. Combined with previous researches, the works of this paper are follows:1. Based on current technologies of dynamic resource allocation in virtual machine and live migration, this thesis designed a framework model of cluster monitoring and management system which divides the cluster into center node and work node that are separately responsible for all nodes’monitoring of cluster and virtual machine monitoring inside the node. The nodes can also adjust automatically when the hotspot appears because of increasing demand.2. Work node monitors the running state of each virtual machine on it in real time. When virtual machine’s application load increase and the initial allocated resources are not enough to meet the demand, the resource utilization rate will reach the threshold followed a warning message. We build a model to forecast load relying on historical records and dynamically allocate resources using the algorithm to guarantee the virtual machine has enough recourses to run application in case of decline of quality of service.3. Center node is in charge of monitoring all nodes in cluster in order to make migration scheme for hot node which cannot ensure its virtual machines’quality of service because of node’s load increase and remaining resources decrease. The model is built based on forecast migration cost with memory size and page modifying speed to select virtual machine to migrate. The load sensitive target node selection algorithm is designed to determine a target host with the state of load and resources to place virtual machine. Reasonable simulation experiments are performed to verify the good effects of virtual machine selection algorithm.更多还原