KVM虚拟机热迁移算法分析及优化

发布时间：2019-02-13 22:02

【摘要】：虚拟化技术自诞生以来就受到人们的关注,由于其在资源利用率、安全性、管理便捷性方面的优势,它被越来越多的企业所采用。虚拟机热迁移技术,是虚拟化技术的一大亮点,它允许在虚拟机运行时进行平滑迁移,迁移过程中服务不中断,因此被广泛应用于负载均衡、能耗控制、服务器在线维护领域。KVM(kernel-based Virtual Machine),一种开源的基于硬件虚拟化的虚拟化软件,由于其简单性、高效性,被越来越多的云厂商用于构建云平台。本文基于KVM虚拟机,深入研究了KVM虚拟机及其热迁移技术的实现。在研究过程中,我们发现现有的KVM热迁移算法存在两个缺点：对脏页的过度重复拷贝和热迁移有时不能终止。针对上述两个缺点,本文提出了两种相应的优化方案,即基于局部性原理的内存拷贝策略优化和自适应降低VCPU热度优化算法：1.基于局部性原理的内存拷贝策略优化。在热迁移的迭代拷贝阶段,需要重复拷贝内存脏页,而程序的运行都具有局部性原理,本轮被修改的内存页下一轮很可能再被修改,因此很多脏页的拷贝其实是无意义的。基于局部性原理的内存拷贝策略优化,根据前两轮的脏页信息,为本轮脏页拷贝提供参考,对于连续两轮写脏的内存页不予以拷贝,从而有效减少了拷贝数据量。实验结果表明,当虚拟机修改内存频繁时,该优化可以显著提高热迁移效率。2.自适应降低VCPU热度优化算法。对于一些内存高负载的虚拟机,如正在执行编译内核任务的虚拟机,由于其脏页产生速率高于内存传输速率,因此迁移可能无法有效终止。对此我们提出了自适应降低VCPU热度的优化算法,通过往VCPU线程周期性插入sleep任务,减少虚拟机运行时间,从而达到降低脏页产生速率的目的,使热迁移顺利结束。实验结果表明,即使在内存高负载情况下,该优化仍然可以顺利结束迁移。
[Abstract]:Since its birth, virtualization technology has attracted people's attention. Because of its advantages in resource utilization, security and management convenience, virtualization technology has been adopted by more and more enterprises. Virtual machine thermal migration technology is one of the highlights of virtualization technology. It allows smooth migration while the virtual machine is running, and the service is not interrupted during the migration process, so it is widely used in load balancing, energy consumption control, Server online maintenance domain. KVM (kernel-based Virtual Machine), an open source virtualization software based on hardware virtualization, is used by more and more cloud vendors to build cloud platform because of its simplicity and efficiency. Based on KVM virtual machine, the realization of KVM virtual machine and its heat transfer technology is studied in this paper. In the course of the research, we find that the existing KVM heat transfer algorithms have two disadvantages: the excessive duplication of dirty pages and the fact that the heat transfer can not be terminated. In view of the above two shortcomings, this paper proposes two corresponding optimization schemes, that is, the optimization of memory copy strategy based on the principle of locality and the optimization algorithm of adaptive reduction of VCPU heat: 1. Optimization of memory copy strategy based on locality principle. In the iterative copy phase of heat transfer, it is necessary to duplicate the dirty pages of memory, and the operation of the program has the principle of locality. The next round of the memory pages modified in this round is likely to be modified again, so many copies of dirty pages are actually meaningless. Based on the principle of locality, the memory copy strategy is optimized. According to the first two rounds of dirty page information, this paper provides a reference for the dirty page copy in this round, and does not copy the dirty memory page in two consecutive rounds, thus effectively reducing the amount of copy data. The experimental results show that the optimization can significantly improve the heat transfer efficiency when the virtual machine modifies memory frequently. Adaptive VCPU heat reduction optimization algorithm. For some virtual machines with high memory load, such as those performing the task of compiling the kernel, the migration may not be terminated effectively because the dirty page generation rate is higher than the memory transfer rate. In this paper, we propose an adaptive optimization algorithm to reduce the heat of VCPU. By inserting sleep task periodically into the VCPU thread, we can reduce the running time of virtual machine, so as to reduce the rate of dirty page generation and make the heat transfer end smoothly. The experimental results show that the optimization can finish the migration smoothly even in the case of high memory load.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP302

【相似文献】