面向多目标的多虚拟机内存优化技术研究

发布时间：2018-04-10 10:39

本文选题：虚拟化技术　+ 内存优化　；引自：《东北大学》2013年硕士论文

【摘要】：随着计算机技术的快速发展和计算机应用项目的广泛使用,云计算作为一种创新的计算模式,在科学计算和商业领域都受到了高度重视。大型数据中心提供的服务以及所使用的虚拟化技术构成了云计算的基础架构。虚拟化技术包括对内存资源的虚拟化。上层应用服务并发请求和系统内各种应用程序的动态变化给虚拟化底层内存资源的优化分配策略带来了新的挑战。虚拟化系统中多虚拟机之间缺少高效的动态内存调节机制,由于各虚拟机对内存资源需求的持续改变,使得内存资源在多虚拟机中出现不均衡分配的现象,部分虚拟机内存资源大量空闲,部分虚拟机内存资源严重不足,导致虚拟机的服务性能低下,降低了用户体验,所以面对复杂的服务和应用,如何提高多虚拟机的综合性能和服务质量,是多虚拟机内存优化问题的研究重点。本文在传统虚拟机内存优化方法的研究基础上,提出了一种面向多目标的多虚拟机内存优化技术。通过分析传统内存优化方法的不足,针对多虚拟机的多目标进行整体优化,不再局限于局部目标的最优,之后给出了这种内存优化技术的整体过程。优化过程的研究主要包括两个重点,基于RBF-Markov的虚拟机内存需求预测算法和基于遗传算法的多目标内存优化方法。首先使用RBF神经网络对虚拟机的内存需求进行预测,并进一步通过马尔可夫模型对预测值加以修正,提高虚拟机内存需求预测值的准确度。然后建立多目标优化模型,使用遗传算法对其进行求解,通过气球驱动机制对多虚拟机进行内存资源调整,之后使用基于雷达图的评价方法对多虚拟机的性能进行综合评估。最后本文通过实验对这种内存优化技术的可行性和有效性进行验证。实验结果表明虚拟机内存需求预测算法具有较高的准确性,本文提出的面向多目标的多虚拟机内存优化技术具有一定的可行性和有效性,通过动态优化控制多虚拟机的内存资源,使虚拟机在满足用户服务质量目标的同时提高内存资源利用率。在雷达图评价方法中发现多虚拟机之间的性能差距缩小且多虚拟机的综合性能得到了提高。
[Abstract]:With the rapid development of computer technology and the wide use of computer application projects, cloud computing, as an innovative computing model, has received great attention in the field of scientific computing and business.The services provided by large data centers and the virtualization technologies used constitute the infrastructure of cloud computing.Virtualization technology includes virtualization of memory resources.The concurrent requests of upper application services and the dynamic changes of various applications in the system bring new challenges to the optimal allocation of memory resources at the bottom of virtualization.There is a lack of efficient dynamic memory regulation mechanism between multiple virtual machines in virtualization systems. Due to the continuous change of memory resource requirements of each virtual machine, the uneven allocation of memory resources occurs in multiple virtual machines.Some virtual machine memory resources are idle and some virtual machine memory resources are seriously insufficient, which leads to the low service performance of virtual machines and reduces the user experience, so it is faced with complex services and applications.How to improve the comprehensive performance and quality of service (QoS) of multiple virtual machines (MVM) is the focus of memory optimization.Based on the research of traditional virtual machine memory optimization method, a multi-objective multi-virtual machine memory optimization technique is proposed in this paper.By analyzing the shortcomings of the traditional memory optimization methods, the global optimization of multi-objective multi-virtual machines is not limited to the optimization of local targets, and the overall process of this memory optimization technique is given.The research on the optimization process includes two main points: the virtual machine memory requirement prediction algorithm based on RBF-Markov and the multi-objective memory optimization method based on genetic algorithm.First, RBF neural network is used to predict the memory requirement of virtual machine, and then the prediction value is modified by Markov model to improve the accuracy of the prediction value of virtual machine memory requirement.Then the multi-objective optimization model is established, the genetic algorithm is used to solve it, the memory resource of the multi-virtual machine is adjusted by balloon drive mechanism, and the performance of the multi-virtual machine is evaluated synthetically by using the evaluation method based on radar graph.Finally, the feasibility and effectiveness of this memory optimization technique are verified by experiments.The experimental results show that the virtual machine memory requirement prediction algorithm has high accuracy, and the multi-objective multi-virtual machine memory optimization technology proposed in this paper is feasible and effective.By dynamically optimizing and controlling the memory resources of multiple virtual machines, the virtual machine can improve the utilization of memory resources while satisfying the quality of service goals of users.It is found that the performance gap between multiple virtual machines is narrowed and the comprehensive performance of multiple virtual machines is improved.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP302

【相似文献】