云数据中心的能耗资源调度策略研究

发布时间：2018-05-17 20:03

本文选题：云数据中心 + 能耗　；参考：《电子科技大学》2016年硕士论文

【摘要】：随着云计算的应用和发展,数据中心规模的扩大,随之而来的是数据中心的能耗、资源利用率等问题日益突出。因此,设计高效的资源分配策略,提高数据中心的资源利用率,降低数据中心的能耗成为一个研究的热点。而能耗作为资源调度的指标,首先会受到服务器性能的影响,其次也会受到用户应用需求的影响。在现代数据中心中,虚拟化技术被广泛应用,但因为虚拟机资源分配被证明是NP难问题,又因为各种应用业务需求不一样以及物理环境的多样性都给能耗计算研究带来了挑战,尤其在全球能源紧缺和温室效应逐年增强的背景下,节能调度成为值得深入研究的课题。本论文从现有的资源监控系统和能耗模型两方面对能耗问题进行研究和分析,通过在每个服务器上放置一个节点代理来获取其资源使用情况。然后根据工程实验得出服务器在不同CPU利用率下的功耗值,计算出能耗模型的参数。根据现有的理论研究了虚拟机迁移准则和触发机制以及待迁移服务器的虚拟机选择策略。然后比较三种选择策略机制,哪一种更能适应实验室这种小规模的云数据中心,得出了最少选择策略相比较而言最能降低其能耗。本论文针对多个虚拟机的放置问题,建立了以负载均衡为目标的优化函数,该优化函数的最优解将使得所有物理服务器的平均利用率保持在一个期望的最优值附近,同时最优解也描述了最优的虚拟机放置策略。为了求解该优化模型,设计了相应的遗传算法,染色体采用二进制编码形式,并采用了随机选择、两点交叉、精英原则等方法来实现遗传算法。然后和工程项目中经常用到的装箱算法和随机放置算法得到的放置序列求得适应度函数值进行比较,结果证明遗传算法求解的最优放置策略可以使得服务器CPU利用率最接近于期望的最优值,从而达到最好的负载均衡效果,负载均衡同时也意味着可以有效减少所需服务器的数量,从而达到降低能耗的作用。最后研究了云计算系统下的节能高效调度机制,建立了云计算系统的性能评估模型和能耗评估模型,并进一步提出了基于利润函数的性能-能耗联合优化函数,然后设计了遗传算法求解最优的请求分发策略和资源分配策略,该最优解意味着一种性能和能耗均衡的调度策略,将比单指标优化更加全面合理。通过实验分析详细描述了系统理论评估模型的分析方法、节能高效调度机制的运行过程以及在系统利润上取得的优化效果。
[Abstract]:With the application and development of cloud computing, the expansion of data center scale, followed by data center energy consumption, resource utilization and other issues become increasingly prominent. Therefore, designing efficient resource allocation strategy, improving resource utilization of data center and reducing energy consumption of data center has become a hot research topic. As an indicator of resource scheduling, energy consumption is affected by the performance of the server first, and then by the application requirements of the user. In modern data centers, virtualization technology is widely used, but because the allocation of virtual machine resources is proved to be NP-hard problem, and because of the different business requirements of various applications and the diversity of physical environment, it brings challenges to the research of energy consumption computing. Especially in the background of global energy shortage and Greenhouse Effect increasing year by year, energy saving scheduling becomes a subject worthy of further study. In this paper, the problem of energy consumption is studied and analyzed from the two aspects of resource monitoring system and energy consumption model, and the resource usage is obtained by placing a node agent on each server. Then, according to the engineering experiment, the power consumption of the server under different CPU utilization is obtained, and the parameters of the energy consumption model are calculated. According to the existing theories, the migration criteria and trigger mechanism of virtual machine and the virtual machine selection strategy of server to be migrated are studied. Then compared with the three selection strategy mechanisms which is more suitable for the small scale cloud data center such as the laboratory the least choice strategy is the most effective to reduce the energy consumption compared with the least choice strategy. In this paper, an optimization function aiming at load balancing is established for the placement of multiple virtual machines. The optimal solution of the optimization function will keep the average utilization of all physical servers near a desired optimal value. At the same time, the optimal solution also describes the optimal virtual machine placement strategy. In order to solve the optimization model, the corresponding genetic algorithm is designed, the chromosome is coded in binary form, and the methods of random selection, two-point crossover and elitist principle are used to realize the genetic algorithm. Then compared with the packing algorithm often used in engineering projects and the placement sequence obtained by the random placement algorithm, the fitness function values are obtained. The results show that the optimal placement strategy solved by the genetic algorithm can make the server CPU utilization close to the expected optimal value, so as to achieve the best load balancing effect. Load balancing also means reducing the number of servers needed to reduce energy consumption. Finally, the energy-saving and efficient scheduling mechanism in cloud computing system is studied, the performance evaluation model and energy consumption evaluation model of cloud computing system are established, and the joint optimization function of performance-energy consumption based on profit function is proposed. Then genetic algorithm is designed to solve the optimal request distribution strategy and resource allocation strategy. The optimal solution means a scheduling strategy with balanced performance and energy consumption, which is more comprehensive and reasonable than the single index optimization. Through the experimental analysis, the analysis method of the system theory evaluation model, the running process of the energy saving and efficient scheduling mechanism and the optimization effect of the system profit are described in detail.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP18;TP308

【参考文献】