基于脏页率预测的虚拟机动态迁移研究

发布时间：2018-07-16 19:32

【摘要】：虚拟机动态迁移中，内存迁移机制最复杂、传输内容最多，目前多使用预拷贝方法。由于迁移期间虚拟机仍在运行，内存页会发生改变，所以预拷贝中需要多次传输脏页，高脏页的频繁重传导致迁移效率严重下降。如何降低迁移时间、减少数据传输量成为虚拟机动态迁移中迫切需要解决的问题。本文针对虚拟机动态迁移中的内存迁移展开研究。首先，本文提出了一种基于脏页率预测的内存预拷贝方法。根据局部性原理的时间局部性，高脏页在短时间内会被持续改写，具有一定的延续性，满足自回归过程的平稳性条件。因此本文使用基于时间序列的自回归模型AR(n)对虚拟机内存页的脏页率做出预测，从迁移期间产生的脏页中识别出高脏页，在迭代过程中优先传送脏页率低的内存页，以避免高脏页的多次重传，进而获得更高的迁移效率。其次，本文在Xen中实现了基于脏页率预测的虚拟机动态迁移，然后在不同CPU负载、内存负载和网络负载下对虚拟机动态迁移性能做了相应测试，并从总迁移时间、停机时间、迁移传输数据量三个方面对实验结果进行了分析。实验结果表明，基于脏页率预测的虚拟机动态迁移减少了数据传输量，缩短了迁移时间，有效提高了虚拟机动态迁移的效率。最后，本文将虚拟机动态迁移技术应用于自主开发的“信息安全设计型实验平台”，设计并开发了名为ISEP的虚拟机动态迁移系统对实验平台的虚拟机集群进行资源调度。ISEP迁移系统优化了资源的计量方法，引入K/N模型避免瞬时负载峰值造成频繁迁移，并综合最小代价原则和最佳匹配原则选择迁移的源虚拟机和目标宿主机，使实验平台的虚拟机集群同时达到负载均衡和节能目标。
[Abstract]:In the dynamic migration of virtual machines, memory migration mechanism is the most complex and the content of transfer is the most. Since the virtual machine is still running during the migration, the memory pages will change, so it is necessary to transfer dirty pages many times in the precopy. The frequent retransmission of high dirty pages results in a serious decrease in migration efficiency. How to reduce the migration time and reduce the amount of data transfer has become an urgent problem in the dynamic migration of virtual machines. This paper focuses on memory migration in virtual machine dynamic migration. Firstly, a memory precopy method based on dirty page rate prediction is proposed. According to the temporal locality of the principle of locality, the highly dirty pages will be continuously rewritten in a short time, which has a certain continuity and satisfies the stationary condition of the autoregressive process. Therefore, this paper uses AR (n), an autoregressive model based on time series, to predict the dirty page rate of virtual machine memory pages, identify high dirty pages from the dirty pages generated during migration, and give priority to transferring memory pages with low dirty page rate during iteration. To avoid multiple retransmissions of high-dirty pages, thus achieving higher migration efficiency. Secondly, the dynamic migration of virtual machine based on the prediction of dirty page rate is realized in Xen, and the dynamic mobility of virtual machine is tested under different CPU load, memory load and network load. The experimental results are analyzed from three aspects of transferring data. The experimental results show that the dynamic migration of virtual machines based on the prediction of dirty page rate reduces the amount of data transfer, shortens the migration time, and effectively improves the efficiency of the dynamic migration of virtual machines. Finally, the virtual machine dynamic migration technology is applied to the "Information Security Design experiment platform", which is developed by ourselves. A virtual machine dynamic migration system named ISEP is designed and developed to optimize the resource measurement method for the virtual machine cluster of the experimental platform. The K / N model is introduced to avoid frequent migration caused by the instantaneous load peak. Combining the minimum cost principle and the best matching principle, the migrating source virtual machine and the target host are selected, so that the virtual machine cluster of the experimental platform can achieve load balancing and energy saving simultaneously.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【参考文献】