数据中心能效管理多目标优化策略研究

发布时间：2018-01-05 08:01

本文关键词：数据中心能效管理多目标优化策略研究　出处：《吉林财经大学》2017年硕士论文　论文类型：学位论文

【摘要】：在大数据的时代背景下,随着云计算技术在全世界的快速普及和发展,云数据中心的基础设施和相关配套设施的数量也在高速地增长。数据中心大量的计算密集型和数据密集型操作需要快速有效地响应,以保证数据中心的正常运转。海量服务器间的协同配合会产生大量的能源消耗,同时,数据中心对于能源的利用率还待提高,这样就对云数据中心运营成本造成巨大的浪费。因此,云数据中心的能耗问题亟待解决。当前,云数据中心的能耗问题得到了国内外学者的广泛关注,主要的解决策略分为硬件节能和软件节能策略两个方面,在软件节能方面,其中的虚拟化技术已经被证实是解决云数据中心能耗问题的有效途径,也是本文的关注重点。本文主要聚焦于虚拟机选择和虚拟机分配过程。实时虚拟机(VM)整合是提高绿色数据中心能效管理水平的有效方法。目前,绿色数据中心的能耗评估模型是以CPU占用率为主要的影响因素。然而,由于GPU的密集处理产生巨大的能耗,原有的能耗评估模型并不适合于数据密集型计算。在本文中,我们提出了基于CPU和GPU利用率的一种新的能效管理评估模型,并提出两种实时动态迁移虚拟机的策略:一个应用于虚拟机选择,另一个应用于虚拟机分配。一些研究人员已经分别基于VM选择策略或VM分配政策提出了自己的解决方案。然而,将虚拟机选择和虚拟机分配这两个策略集成在一起,将会得到一个更为高效的实时动态迁移的虚拟机整合策略。基于此,一个快速的基于人工蜂群算法(ABC)的实时VM整合策略被提出,并结合适合数据密集型计算的能耗评估模型共同组成DataABC策略。DataABC采用了人工蜂群算法的思想,从而得到一个快速并且具有全局优化特点的虚拟机迁移策略。与其他经典的虚拟机整合策略相比,DataABC的总能耗下降明显。在虚拟机分配过程中,传统的分配策略存在着分配速度难以满足数据密集型作业要求的特点,以及容易陷入局部最优等现象。因此,为了满足数据密集型作业对于响应速度的需要,本文引入梯度下降算法,加快人工蜂群算法搜寻局部最优解的速度,同时引入模拟退火算法,加强人工蜂群算法搜寻全局近似最优解的能力,使空闲节点关闭或者休眠来达到节能的目的,从而减少了能源消耗,提高了资源使用效率,减少了数据中心的运营成本。研究者提出了多种节能策略,例如开/关策略,虚拟机整合策略,DVFS策略等,但是,每种策略都有其实现条件和自身特点,将多种节能策略集成,将更好的实现数据中心节能目标,实现数据中心的可持续发展。
[Abstract]:Under the background of big data era, with the rapid spread and development of cloud computing technology in the world. The number of cloud data centers infrastructure and related supporting facilities is also growing at a high speed. A large number of computation-intensive and data-intensive operations in data centers need to respond quickly and effectively. In order to ensure the normal operation of the data center. The cooperation between the massive servers will produce a large amount of energy consumption, at the same time, the data center for energy utilization still needs to be improved. Therefore, the energy consumption of cloud data center needs to be solved urgently. At present, the energy consumption of cloud data center has been widely concerned by scholars at home and abroad. The main solutions are hardware energy saving and software energy saving. In software energy saving, the virtualization technology has been proved to be an effective way to solve the problem of cloud data center energy consumption. This paper focuses on the selection and allocation of virtual machines. The integration of real-time virtual machines (VMs) is an effective way to improve the energy efficiency management level of green data centers. The energy consumption assessment model of green data center is based on the CPU occupancy rate. However, because of the intensive processing of GPU, the energy consumption is huge. The original energy consumption evaluation model is not suitable for data-intensive computing. In this paper, we propose a new energy efficiency management evaluation model based on CPU and GPU utilization. Two strategies for real-time dynamic migration of virtual machines are proposed: one is applied to virtual machine selection. Another application is virtual machine allocation. Some researchers have proposed their own solutions based on VM selection strategy or VM allocation policy respectively. Integrating the two strategies of virtual machine selection and virtual machine allocation will result in a more efficient virtual machine integration strategy for real-time dynamic migration. A fast real-time VM integration strategy based on artificial bee colony algorithm (ABC) is proposed. Combined with the energy consumption evaluation model suitable for data-intensive computing, DataABC strategy. DataABC adopts the idea of artificial bee colony algorithm. Thus, a fast and globally optimized virtual machine migration strategy is obtained. Compared with other classical virtual machine integration strategies, the total energy consumption of DataABC is significantly reduced, and in the process of virtual machine allocation. The traditional allocation strategy has the characteristics that the allocation speed is difficult to meet the requirements of data-intensive jobs, and it is easy to fall into the local optimum. Therefore, in order to meet the needs of the response speed of data-intensive jobs. In this paper, gradient descent algorithm is introduced to speed up the search speed of artificial bee colony algorithm, and simulated annealing algorithm is introduced to enhance the ability of artificial bee colony algorithm to search global approximate optimal solution. The idle nodes are closed or dormant to achieve the purpose of energy saving, thus reducing energy consumption, improving the efficiency of resource use, and reducing the operating costs of the data center. Researchers have proposed a variety of energy-saving strategies. For example, on / off strategy, virtual machine integration strategy, DVFS strategy, etc., however, each strategy has its own implementation conditions and its own characteristics, the integration of a variety of energy-saving strategies will better achieve the data center energy-saving goals. To realize the sustainable development of data center.
【学位授予单位】：吉林财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308

【参考文献】