面向云数据中心的资源管理机制研究

发布时间：2018-06-21 09:21

本文选题：数据中心 + 资源监控　；参考：《电子科技大学》2017年硕士论文

【摘要】：近年,云计算技术的发展和成熟,使得云计算已经被广泛的应用在各行各业(如军事、教育、金融、电子政务、电商等)中,大量的传统IT系统纷纷迁移到云数据中心,此外,大数据、移动互联网、物联网等的飞速发展和应用,进一步加速了云计算的基础设施也就是云数据中心数量和规模的增长,这也造成了现代云数据中心规模极其庞大、结构异常复杂等特点。除此之外,现代多样的应用需求背景下,资源管理者除了需要关心用户对可靠性、性能等服务质量之外,还要考虑节能减排等问题来降低的开支,实现利润的增大,这给云数据中心的管理者带来了不小的挑战。为了实现云数据中心高效、高可靠和低能耗的资源管理,本文分别从云数据中心资源监控、关联建模和资源调度三个方面进行了研究,核心的贡献点如下:1)研发了基于仿生自主神经系统(Bionic Autonomic Nervous System,BANS)的云资源监控系统。借鉴BANS的设计思想,设计了一种资源监控系统,它充分的考虑现代云数据中心的虚拟化和超大规模等特点。其通过分层式的设计,使得系统的扩展性良好;除此之外,该监控系统还具有BANS的自主性,能够实现某种程度上的自组织、自诊断、自修复和自优化,进而减轻监控系统主节点的负载,使得监控系统可以更好的适用于现代的大规模云数据中心。2)提出了一种可靠性、性能和能耗的多维关联模型。其首先在考虑硬件随机失效的前提下,分别建立了可靠性-性能和可靠性-能耗两个关联子模型,实现了更加准确合理的云服务的性能和能耗分析;然后通过利润模型——利用时间功效函数评估性能的收益,电能消耗来评估支出——实现了可靠性、性能和能耗三者的关联分析,为现代云数据中心可靠性、性能和能耗评估和调度管理建立了模型基础。3)提出了一种自优化的资源管理框架。充分考虑了VM对资源需求是动态变化的这一前提,本文借鉴BANS的思想,提出了一种基于BANS的云资源管理系统用于云资源动态管理,其将监控系统获取的监控信息作为基础,通过系统各个模块的配合实现了一种自优化的动态云资源管理框架。其通过虚拟机的迁移来满足虚拟机对资源的动态需求:当物理机出现过载时,可以选择迁出虚拟机,减少不必要服务性能退化;而当物理机欠载时,还可以通过迁移整合虚拟机来减少开启的物理机总数,进而提高物理机资源的利用率。4)提出了可靠性感知的多数据中心能耗成本建模方法。在考虑可靠性的基础上,建立分布式多数据中心的系统、任务、调度、功耗和约束模型,将多数据中心管理问题公式化为了满足用户对性能、可靠性约束的前提下,最小化多数据中心能耗成本的问题,并提出可靠性感知的分布式数据中心能耗成本优化的调度算法,其在考虑可靠性的基础上,充分利用了不同地域数据中心电价的差异性,实现跨域的多数据中心能耗成本优化。
[Abstract]:In recent years, with the development and maturity of cloud computing technology, cloud computing has been widely used in all walks of life (such as military, education, finance, e-government, e-commerce and so on). A large number of traditional IT systems have migrated to cloud data centers. With the rapid development and application of big data, mobile Internet, Internet of things and so on, cloud computing infrastructure, that is, the number and scale of cloud data centers, has been further accelerated, which has resulted in the extremely large scale of modern cloud data centers. The structure is extremely complex and so on. In addition, in the context of modern and diverse application requirements, resource managers need to care about the reliability, performance and other service quality of the user, but also consider energy saving and emission reduction to reduce the expenditure and realize the increase of profits. This poses a great challenge to the management of cloud data centers. In order to realize high efficiency, high reliability and low energy consumption resource management of cloud data center, this paper studies three aspects of cloud data center resource monitoring, association modeling and resource scheduling, respectively. The core contribution is as follows: 1) A cloud resource monitoring system based on Bionic Autonomic Nervous system is developed. Based on the design idea of Bans, a resource monitoring system is designed, which fully considers the virtualization and large scale of modern cloud data center. In addition, the monitoring system has the autonomy of Bans, which can realize some degree of self-organization, self-diagnosis, self-repair and self-optimization. Then the load of the master node of the monitoring system is reduced, which makes the monitoring system more suitable for the modern large-scale cloud data center. 2) A multi-dimensional correlation model of reliability, performance and energy consumption is proposed. Firstly, two correlation sub-models of reliability performance and reliability energy consumption are established on the premise of hardware random failure, which realizes more accurate and reasonable analysis of cloud service performance and energy consumption. Then through the profit model-using the time efficiency function to evaluate the performance income, the electric energy consumption to evaluate the expenditure-to realize the reliability, the performance and the energy consumption correlation analysis, for the modern cloud data center reliability, The performance and energy consumption evaluation and scheduling management are modeled on the basis of. 3) A self-optimizing resource management framework is proposed. Considering the premise that VM is a dynamic change of resource requirement, this paper proposes a cloud resource management system based on banks, which takes the monitoring information obtained by monitoring system as the basis, and proposes a new cloud resource management system based on Bans. A self-optimizing dynamic cloud resource management framework is implemented through the cooperation of various modules of the system. The migration of virtual machine can satisfy the dynamic resource demand of virtual machine: when the physical machine is overloaded, it can choose to move out of the virtual machine to reduce the unnecessary service performance degradation, and when the physical machine is under load, It is also possible to reduce the total number of open physical machines by migrating and integrating virtual machines, and then improve the utilization ratio of physical computers. 4) A reliability aware multi-data center energy cost modeling method is proposed. On the basis of considering reliability, the system, task, scheduling, power consumption and constraint model of distributed multi-data center are established, and the multi-data center management problem is formulated to satisfy the performance and reliability constraints of users. The problem of minimizing the cost of multi-data center energy consumption is minimized, and a distributed data center cost optimization scheduling algorithm based on reliability perception is proposed. On the basis of considering reliability, it makes full use of the difference of electricity price between different regional data centers. The energy cost optimization of multi-data center across domains is realized.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308

【参考文献】