基于虚拟集群在线迁移的虚拟化资源整合算法研究

发布时间：2018-03-15 01:16

本文选题：云计算　切入点：虚拟集群迁移　出处：《吉林大学》2014年硕士论文　论文类型：学位论文

【摘要】：近30年来，随着系统虚拟化和网络虚拟化技术不断地发展，虚拟化技术所呈现的隔离性和安全性以及可定制的特征，越来越得到人们的信赖，其本身存在的问题也得到了很好地解决，预拷贝，数据包缓存等技术，降低了虚拟机迁移的成本，推动了虚拟机迁移技术的广泛应用。例如，应用虚拟化技术的集群/云计算环境具有资源利用率高，方便管理维护，可靠性高等特点，可灵活地部署和管理高性能计算环境——虚拟集群。虚拟化技术为系统资源管理提供了灵活的手段，其中虚拟机迁移技术在负载平衡，系统容错，在线维护，能耗管理等资源管理调度领域发挥着重大作用。但在现实环境中，想要真正地利用好虚拟机迁移技术进行资源的动态管理，必须了解工作负载本身的特点，才能获得收益。例如，web服务的高并发性能及网络传输性能需求较高，而在虚拟集群中运行的并行作业，初始时当所有子任务就绪时才能开始进行，计算过程中，子任务间同步性能需求很高，关联性强，且有的作业子任务间数据交互密集，而且当所有子任务完全结束时，并行作业才算完成，甚至一起释放资源。这些特点均需要管理员虑及，，才能在保证用户性能需求的前提下，合理提高系统利用效率。现阶段的研究工作大多单纯地以资源利用率为目标进行虚拟机部署迁移，在用于部署虚拟集群的集群／云计算环境中时，很容易造成资源碎片，难以直接应用到虚拟集群环境中。综上，本文从运行并行作业的虚拟集群出发，对虚拟机迁移策略进行重新思考，在以减少资源碎片为目的的虚拟化资源整合的应用场景中，制定了相应的虚拟机迁移策略。关注了集群／云计算环境本身的特点，及并行作业子任务间交互性强等特点。首先对算法实施的环境进行系统模型构造，进行了合理的假设说明。然后利用“区域划分法”对异构环境下的碎片最小化问题的规模进行缩减，显著降低了问题的维度和难度，提高了算法的可实施性。最后提出了在同构子域这个子环境中，以虚拟集群为轴心的虚拟机迁移策略。在试验阶段，对系统资源现状的原因进行简要分析，采用同种负载，多种资源分配方式的模拟方法模拟现实中的资源分配使用现状。通过参数调节，可以很好地模拟真实数据。理论上以虚拟集群为轴心的虚拟机迁移策略，虚拟集群也“紧凑”，系统资源的分配释放越整齐有序，结果有利于虚拟化资源整合。
[Abstract]:In the past 30 years, with the continuous development of system virtualization and network virtualization technology, the isolation, security and customizable characteristics of virtualization technology have gained more and more people's trust. The existing problems have been well solved, such as precopy, packet cache and so on, which reduce the cost of virtual machine migration and promote the wide application of virtual machine migration technology. For example, The cluster / cloud computing environment with virtualization technology has the advantages of high resource utilization, convenient management and maintenance, high reliability and so on. It can flexibly deploy and manage the high performance computing environment-virtual cluster. Virtualization technology provides flexible means for system resource management, in which virtual machine migration technology plays an important role in the field of resource management scheduling, such as load balancing, system fault tolerance, online maintenance, energy consumption management, etc. In order to make good use of virtual machine migration technology for dynamic resource management, we must understand the characteristics of the workload itself in order to gain benefits. For example, the high concurrent performance of web services and the high demand for network transport performance, The parallel jobs running in the virtual cluster can only start when all the sub-tasks are ready. In the computing process, the performance of synchronization among the sub-tasks is very high, the correlation is strong, and the data interaction between the sub-tasks of some jobs is dense. And when all the subtasks are complete, the parallel jobs are done, and even the resources are released together. These characteristics require the administrator to take into account the performance requirements of the user. At present, most of the research work only aims at resource utilization to deploy and migrate virtual machines. In the cluster / cloud computing environment which is used to deploy virtual clusters, it is easy to cause resource fragmentation. It is difficult to directly apply to virtual cluster environment. In summary, this paper reconsiders the strategy of virtual machine migration from the virtual cluster running parallel jobs, in the application scenario of virtual resource integration, which aims to reduce resource fragmentation. The migration strategy of virtual machine is formulated. The characteristics of cluster / cloud computing environment and the interactivity among parallel task subtasks are concerned. Firstly, the system model of the algorithm implementation environment is constructed. The reasonable hypothesis is given, and then the scale of the debris minimization problem in heterogeneous environment is reduced by using the "region partition method", which reduces the dimension and difficulty of the problem. Finally, a virtual machine migration strategy with virtual cluster as its axis in the isomorphic sub-domain is proposed. In the experiment stage, the reasons of the current system resources are analyzed briefly, and the same load is adopted. The simulation method of multiple resource allocation methods simulates the current situation of resource allocation in reality. By adjusting parameters, real data can be well simulated. In theory, virtual machine migration strategy with virtual cluster as its axis can be used. Virtual cluster is also "compact", the more neat and orderly the distribution of system resources, the more conducive to the integration of virtualized resources.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.01

【参考文献】