虚拟机整合若干关键问题研究

发布时间：2018-04-14 23:10

本文选题：虚拟化 + 数据格式　；参考：《西北师范大学》2013年硕士论文

【摘要】：为了提高数据中心所有物理服务器的资源利用率和能源使用率，可以通过动态整合虚拟机的方式来实现。虚拟机动态整合的主要实现方式有物理服务器之间虚拟机的实时迁移和动态地将空闲的物理服务器转换到低能耗模式。本文主要研究监控数据的传输格式，资源利用预测以及物理服务器超载和低负载情况下的虚拟机重定位。首先，分析典型监控系统的数据传输格式，结合监控信息的特点，提出一种优化的数据传输格式，该格式将监控信息进行统一的管理。其次，通过分析灰色模型和马尔科夫理论的特点，提出自反馈灰色马尔科夫预测模型(Self-feedback Grey Markov model，SfGM)，该模型以每台虚拟机的某一资源利用的N个连续的历史数据作为输入，通过分析这N个数据的内在关系，预测虚拟机未来一段时间对该资源的利用情况。最后，动态的监控数据中心的所有虚拟机和物理服务器的资源占用情况和运行状态，当存在虚拟机或物理服务器满足预先设定的整合条件时，调用SfGM来预测资源的使用情况，通过预测结果来判断在当前时间点是否需要整合虚拟机。当需要执行虚拟机整合时，利用本文提出的最重物理机最适合虚拟机优先重定位(the Heaviest PM the mostSuitable VM First Relocation， HPSVFR)算法进行虚拟机的重定位。仿真实验表明，优化的数据格式能够有效减少传输的信息量，节约网络带宽。SfGM的预测具有最高的准确性，最好情况下其预测值数学期望的相对误差是GMM的46.8%，GM(1,1)的28.6%。HPSVFR算法与最先适应和最好适应算法相比较，重定位开销最少，，仅为它们的70%左右。本文中的虚拟机整合架构能够有效地判断当前数据中心的资源使用情况，如果当前数据中心存在处于超载或者低负载状态的物理服务器和虚拟机时，能够有效的利用HPSVFR算法实现虚拟机的重定位。
[Abstract]:In order to improve the resource utilization and energy utilization of all physical servers in the data center, the virtual machine can be dynamically integrated.The main ways to realize the dynamic integration of virtual machines are the real-time migration of virtual machines between physical servers and the dynamic conversion of idle physical servers to low energy consumption mode.This paper focuses on the transmission format of monitoring data, resource utilization prediction and virtual machine repositioning under physical server overload and low load.First of all, the data transmission format of typical monitoring system is analyzed, and an optimized data transmission format is proposed according to the characteristics of monitoring information. This format unified the management of monitoring information.Secondly, by analyzing the characteristics of grey model and Markov theory, a self-feedback grey Markov prediction model, self-feedback Grey Markov model SfGMN, is proposed. The model takes N consecutive historical data from a resource of each virtual machine as input.By analyzing the internal relations of these N data, the utilization of this resource in the future is predicted.Finally, it dynamically monitors the resource occupation and running status of all virtual machines and physical servers in the data center. When there is a virtual machine or physical server that meets the pre-set integration conditions, SfGM is called to predict the use of resources.The prediction results are used to determine whether or not the virtual machine needs to be integrated at the current point in time.When it is necessary to perform virtual machine integration, the Heaviest PM the mostSuitable VM First repositioning (HPSVFR) algorithm proposed in this paper is used to relocate the virtual machine.The simulation results show that the optimized data format can effectively reduce the amount of information transmitted and save the network bandwidth. The prediction of SfGM has the highest accuracy.In the best case, the relative error of the mathematical expectation of its prediction value is 46.8% of GMM. Compared with the first adaptive algorithm and the best adaptive algorithm, the repositioning cost of the 28.6%.HPSVFR algorithm is the least, which is only about 70% of that of the first adaptive algorithm and the best adaptive algorithm.The virtual machine integration architecture in this paper can effectively judge the resource usage of the current data center, if the current data center has physical servers and virtual machines in an overloaded or low-load state,Can effectively use the HPSVFR algorithm to achieve virtual machine relocation.
【学位授予单位】：西北师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【参考文献】