云数据中心中虚拟机放置和实时迁移研究

发布时间：2018-04-20 15:11

本文选题：云计算 + 数据中心　；参考：《北京交通大学》2013年博士论文

【摘要】：云计算是一种新的商业计算模型和服务模式,它将计算任务分布在大量计算机构成的资源池上,使用户能够按需获取计算能力、存储空间和信息服务。目前,云计算已经在网络搜索、科学计算、虚拟环境、能源和生物信息等领域开始了相关的应用和探索。随着云计算的发展,云计算基础设施的规模正在急剧地变大。云计算数据中心通常拥有成千上万台物理结点,在其上运行着各种各样的服务和应用程序。在这些系统中,具有不同资源请求和动态工作负载的各种应用程序共享着数据中心的资源。虚拟化技术则通过允许计算系统按需分配资源和动态迁移工作负载的方式来支持数据中心的资源共享。然而,由系统规模的增加带来系统管理的复杂性和系统内在的动态性,对云计算数据中心中的虚拟机资源管理来说具有很大的挑战性。本文在全面地分析云计算技术和数据中心资源管理研究现状的基础上,重点对云计算数据中心中虚拟机的放置和实时迁移展开了相关研究,并取得了一些创新和成果。本文的主要研究内容及创新点为： 1、提出多目标优化的虚拟机初始化放置策略。首先给出云计算基础设施中虚拟机放置的管理框架,并将虚拟机初始化放置问题定义为一个多目标优化问题。然后提出基于最大最小蚂蚁系统和排除法的虚拟机放置问题的多目标优化蚁群算法。该算法是一种分布式优化方法,有利于并行计算；并且具有正反馈机制,通过信息素的不断更新可以高效地收敛到最优解。实验结果证明,与传统的启发式方法和遗传算法相比,提出的多目标优化虚拟机放置蚁群算法能够在多个相互冲突的目标之间实现最优权衡和折衷,使得在发生SLA违背概率较少的情况下,系统的资源浪费和电源消耗也较少。 2、提出多目标优化的虚拟机动态管理方法。在管理过程中,系统考虑提高用户服务质量、减少资源竞争和降低电源消耗等3个目标。管理方法主要解决由于系统条件的改变或应用负载的动态变化所引起的虚拟机重新分配问题,即决策什么时候进行虚拟机迁移、迁移哪些虚拟机和被迁移的虚拟机放置到哪些物理结点上。在决定何时进行虚拟机迁移的问题上,系统针对不同的优化目标设置高低门限值,并采用滑动窗口和时间序列趋势预测法来判断是否触发虚拟机的迁移；在选择哪些虚拟机进行迁移的问题上,系统根据不同的优化目标分别采取不同的策略来选择虚拟机进行迁移；在选择将虚拟机迁移到哪些物理结点的问题上,系统采用解决多准则决策分析方法中的TOPSIS方法来选择物理结点,目的是为了均衡系统中多个目标之间的冲突。实验结果证明,采用TOPSIS方法的多目标优化管理方法能够得到较低的SLA违背率,较小的资源负载和较少的电源消耗,并且很好地平衡了不同目标之间的冲突。 3、提出一种虚拟机快速实时迁移方法。首先通过对预拷贝方法模型进行公式化描述和性能分析,在预拷贝方法中增加用于标记改动频繁内存页的页位图。然后在迭代过程中采用时间序列方法对内存页进行判断：对于在本轮变脏,并且在之前的迭代中变脏的次数超过给定次数的内存页,则标记为改动频繁的内存页。这些标记过的内存页将只会在最后一轮迭代中传输,从而避免了内存页的反复重传。实验结果证明,与预拷贝方法相比,改进后的预拷贝方法明显地减少了迁移时间和传输数据,达到了快速迁移虚拟机的目的。
[Abstract]:Cloud computing is a new business computing model and service model. It distributes computing tasks on a large pool of resources made up of computers. Users can obtain computing power, storage space and information services on demand. At present, cloud computing has begun to relate in the fields of network search, scientific computing, virtual environment, energy and biological information. Application and exploration.
With the development of cloud computing, the size of cloud computing infrastructure is dramatically increasing. Cloud computing data centers usually have thousands of physical nodes that run a variety of services and applications. In these systems, various applications with different resource requests and dynamic work loads share data centers. The virtualization technology supports resource sharing in the data center by allowing computing systems to allocate resources on demand and dynamically migrating work loads. However, the complexity of system management and the inherent dynamics of the system are brought by the increase of system scale, which is very important to the virtual machine resource management in the cloud computing data center. A big challenge.
On the basis of comprehensive analysis of the current situation of cloud computing technology and data center resource management, this paper focuses on the research on the placement and real-time migration of virtual machines in the cloud computing data center, and has made some innovations and achievements. The main contents and innovation points of this paper are as follows:
1, a virtual machine initialization placement strategy for multi-objective optimization is proposed. First, the management framework of virtual machines in cloud computing infrastructure is given, and the initialization placement problem of virtual machines is defined as a multi-objective optimization problem. Then a multi-objective optimization ant colony based on the maximum minimum ant system and the exclusion method for virtual machine placement is proposed. The algorithm is a distributed optimization method, which is beneficial to parallel computing, and has a positive feedback mechanism, which can efficiently converge to the optimal solution through the continuous updating of pheromone. The experimental results show that the proposed multi target optimization virtual machine placement ant colony algorithm can be more than the traditional heuristic method and genetic algorithm. The best trade-off and compromise between conflicting objectives make the system waste of resources and power consumption less when SLA violation probability is less.
2, a virtual machine dynamic management method for multi-objective optimization is proposed. In the process of management, the system considers 3 goals, such as improving the quality of the user's service, reducing the resource competition and reducing the power consumption. The management method mainly solves the problem of virtual machine redistribution caused by the change of the system conditions or the dynamic changes of the applied load, that is, what to make the decision When the virtual machine and the migrated virtual machine are placed on what physical nodes are migrated by virtual machines, the system sets high and low threshold values for different optimization targets on the issue of deciding when to carry out virtual machines, and uses sliding window and time sequence potential prediction to determine whether the migration of virtual machines can be triggered. In choosing which virtual machines to migrate, the system adopts different strategies to select virtual machines to migrate according to the different optimization objectives. On the problem of selecting the physical nodes to migrate to the virtual machine, the system uses the TOPSIS method to select the physical nodes in the multi criteria decision analysis method. In order to balance the conflicts between multiple targets in the system, the experimental results show that the multi-objective optimization management method using the TOPSIS method can get lower SLA violation rate, smaller resource load and less power consumption, and well balance the impulse between different targets.
3, a fast real time migration method for virtual machines is proposed. First, the precopy method model is formulated with formula description and performance analysis. In the pre copy method, the page bitmap, which is used to change frequent memory pages, is added to the pre copy method. In previous iterations, the number of dirty pages more than a given number of memory pages is marked as a frequently changed memory page. The marked memory pages will only be transmitted in the last round of iterations to avoid repeated retransmission of the memory page. Experimental results show that the improved copy method is significantly reduced compared with the pre copy method. The migration time and data are transferred to achieve the goal of fast migration of the virtual machine.

【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TP308

【相似文献】