Xen虚拟机内存实时迁移技术研究与实现

发布时间：2018-03-31 14:45

本文选题：实时迁移　切入点：虚拟机　出处：《东北大学》2012年硕士论文

【摘要】：虚拟机实时迁移是虚拟化技术的一个重要特征,对于近年来不断兴起的数据中心的负载均衡和灾难恢复有非常重大的意义。通过虚拟机的实时迁移,把虚拟机从源物理主机硬件平台迁移到目的物理主机硬件平台上,不仅可以快速消除源物理机上虚拟机之间的资源竞争,还可以通过合并操作大大减少激活的物理机数目,显著提高计算的效率。目前虚拟机实时迁移大多采用迭代预拷贝方法,该方法在虚拟机低负载或网络环境较好的情况下有良好的表现,但也具有明显的局限性。当脏页率过大时,会产生大量内存页重传,导致迁移期间数据传送总量变大和总迁移时间变长,影响迁移期间虚拟机的服务质量；当虚拟机网络负载过重时,会出现脏页率大于迁移所用网络传输速率的情况,预拷贝算法会失效,导致停机时间过长。本文对Xen虚拟机内存实时迁移技术进行了深入地研究,针对原有迭代预拷贝算法在特定环境下存在的总迁移时间和停机时间过长、迁移期间服务质量降低、算法失效等问题,提出了一种内存实时压缩迁移算法。该算法在预拷贝算法的基础之上,通过设计并实现内存实时压缩算法M2LZO和内存页重复度检查算法,在内存批量传送出去前,先使用M2LZO算法对内存页以不同的重复度作区分,分别使用适合的压缩等级进行压缩编码然后再传送；在目的物理主机接收到压缩数据后先用相应的译码算法进行解压缩,再交给预拷贝算法恢复内存页。该算法可以有效减少迁移期间内存传送总量,缩短总迁移时间和停机时间,而且在网络环境较差的情况下也能表现出良好的适应性,一定程度上解决了原算法瓶颈问题。在此基础上,本文还对Xen迁移管理框架进行了优化。针对原框架不能自动迁移,设计了信息收集器、热点检测和迁移策略等机制；针对原框架存在的性能提升空间,设计了迁移算法的多线程并行处理和内存空闲页特殊处理等方法提升迁移性能；针对迁移期间服务质量的保证,对迁移过程中占用资源情况进行控制,使迁移本身占用的CPU和网络带宽资源达到最优利用。本文最后通过实验对M2LZO算法和优化后的迁移管理框架进行性能测试。实验结果表明,优化后的迁移管理框架与优化前相比,对不同负载状态下虚拟机的迁移性能有不同程度的提升。
[Abstract]:Real-time migration of virtual machines is an important feature of virtualization technology, which is of great significance to load balancing and disaster recovery of data centers that have been emerging in recent years.Through the real-time migration of virtual machines, the virtual machines are migrated from the source physical host hardware platform to the destination physical host hardware platform, which not only can eliminate the resource competition between the source physical machine virtual machines quickly.The number of activated physical machines can be greatly reduced by combining operation, and the efficiency of calculation can be improved significantly.At present, the virtual machine real-time migration mostly uses iterative precopy method, this method has good performance in the case of low load virtual machine or good network environment, but also has obvious limitations.When the dirty page rate is too high, a large number of memory pages will be retransmitted, which will result in a large amount of data transfer during migration and a longer total migration time, which will affect the quality of service of virtual machines during migration; when the network of virtual machines is overloaded, the quality of service of virtual machines will be affected.When the dirty page rate is higher than the transfer rate of the migrated network, the precopy algorithm will fail and the downtime will be too long.In this paper, the memory real-time migration technology of Xen virtual machine is deeply studied. Aiming at the problems of the total migration time and downtime of the original iterative precopy algorithm in a specific environment, the quality of service during migration, the failure of the algorithm, etc.A real-time memory compression migration algorithm is proposed.On the basis of precopy algorithm, this algorithm designs and implements the real-time memory compression algorithm M2LZO and the memory page repetition checking algorithm. Before the memory is sent out in batches, the M2LZO algorithm is used to distinguish the memory pages with different repetitions.After receiving the compressed data, the corresponding decoding algorithm is used to decompress it, and then the precopy algorithm is given to restore the memory page.This algorithm can effectively reduce the total amount of memory transfer during the migration period, shorten the total migration time and downtime, and show good adaptability in the case of poor network environment, to a certain extent, solve the bottleneck problem of the original algorithm.On this basis, this paper also optimizes the Xen migration management framework.For the original frame can not be migrated automatically, the information collector, hot spot detection and migration strategy are designed, and the performance improvement space of the original framework is proposed.Multithread parallel processing of migration algorithm and special processing of memory free pages are designed to improve migration performance, aiming at the guarantee of quality of service during migration, the use of resources during migration is controlled.The optimal utilization of CPU and network bandwidth resources is achieved by migration itself.Finally, the M2LZO algorithm and the optimized migration management framework are tested by experiments.The experimental results show that the migration performance of the virtual machine under different load states is improved by the optimized migration management framework.
【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】