绿色数据中心中的平均温度感知资源控制算法研究

发布时间：2018-05-17 13:17

本文选题：数据中心 + 能耗最小化　；参考：《内蒙古工业大学》2017年硕士论文

【摘要】：短短数年间,云计算技术从提出概念转向大规模应用。云计算技术可以与多种行业进行融合,为用户提供便捷的服务,体现出巨大的应用价值和发展前景。为了满足对计算日益增长的需求,云服务提供商开始运营不同规模的数据中心。与此同时,数据中心消耗的能源也越来越多,逐渐成为制约云服务提供商发展的瓶颈,数据中心的能耗问题已经吸引了工业界和学术界的关注,已成为研究的热点。因此,以最小化数据中心能耗为目标的资源控制策略是本文的主要研究内容。首先,本文对数据中心的能耗最小化问题研究现状进行调研,主要包括数据中心能耗的分布和热循环过程,服务器系统能耗研究现状,制冷系统能耗研究现状,以及目前数据中心能耗最小化的技术。同时,本文还对数据中心能耗动态优化的理论——李雅普诺夫优化理论进行了研究,李雅普诺夫优化理论中的离散时间队列理论和偏移惩罚函数是研究的重点。队列的稳定性理论为期望均值约束条件提供了保证,李雅普诺夫偏移惩罚函数将队列与成本函数相结合,使目标函数找到性能和成本之间的平衡点。根据李雅普诺夫优化理论设计的优化算法能够克服动态优化的不确定性与其求解算法计算复杂度高的不足,可以快速而准确地选取决策变量,同时满足约束条件,适用于优化实地数据中心的能源消耗。其次,本文对数据中心能耗最小化问题的模型进行构建,模型包括服务器能耗模型和制冷系统能耗模型,并形式化了服务质量约束和服务器CPU平均温度约束。以克服已有研究工作多数致力于仅降低服务器系统的能耗而忽略机房制冷系统的能耗,或没有考虑到服务质量(QoS,Quality of Service)约束和服务器CPU温度约束的不足。然后,结合李雅普诺夫优化理论中虚拟队列的概念,本文推演了李雅普诺夫偏移函数的上界,并利用数据中心能耗最小化李雅普诺夫函数设计了线性控制策略和二次控制策略两种能耗最小化控制策略。最后,本文使用数据中心真实工作负载数据进行仿真,得到各项指标的情况,并将设计的能耗最小化控制策略与基准策略进行比较,从数据和理论两个方面对能耗最小化控制策略的性能进行评价,证明这两种控制策略可以实现数据中心总能耗的最小化。并对实验结果做出讨论,给出数据中心温度感知能耗最小化问题的最优解决方案。
[Abstract]:Within a few short years, cloud computing technology from the concept to large-scale applications. Cloud computing technology can be combined with a variety of industries to provide users with convenient services, reflecting a huge application value and development prospects. To meet the growing demand for computing, cloud service providers are operating data centers of different sizes. At the same time, more and more energy is consumed in the data center, which gradually becomes the bottleneck restricting the development of cloud service providers. The energy consumption of the data center has attracted the attention of industry and academia, and has become a hot research topic. Therefore, resource control strategy aiming at minimizing data center energy consumption is the main content of this paper. Firstly, this paper investigates the energy consumption minimization of data center, including the distribution of data center energy consumption and the process of thermal cycle, the research status of energy consumption in server system, and the research status of energy consumption in refrigeration system. And the current data center energy consumption minimization technology. At the same time, the Lyapunov optimization theory, which is the dynamic optimization theory of data center energy consumption, is also studied. The discrete time queue theory and offset penalty function in Lyapunov optimization theory are the key points of the research. The stability theory of the queue guarantees the expected mean constraint. The Lyapunov offset penalty function combines the queue with the cost function to make the objective function find the balance between performance and cost. The optimization algorithm designed according to Lyapunov optimization theory can overcome the uncertainty of dynamic optimization and the high computational complexity of its solution algorithm, and it can select the decision variables quickly and accurately, and satisfy the constraint conditions at the same time. Suitable for optimizing energy consumption in field data centers. Secondly, this paper constructs the model of data center energy minimization, including server energy consumption model and refrigeration system energy consumption model, and formalizes the QoS constraints and server CPU average temperature constraints. In order to overcome the shortcomings of most of the previous researches which only reduce the energy consumption of the server system and neglect the energy consumption of the refrigeration system in the computer room, or do not consider the QoS quality of Service constraint and the CPU temperature constraint of the server. Then, combined with the concept of virtual queue in Lyapunov optimization theory, the upper bound of Lyapunov migration function is deduced. The linear control strategy and the quadratic control strategy are designed by using the Lyapunov function to minimize the energy consumption of the data center. Finally, we use the real workload data of the data center to simulate, get the situation of each index, and compare the designed energy consumption minimization control strategy with the reference strategy. The performance of the energy consumption minimization control strategy is evaluated in terms of data and theory, and it is proved that these two control strategies can minimize the total energy consumption of the data center. The experimental results are discussed and the optimal solution to the problem of data center temperature sensing energy minimization is given.
【学位授予单位】：内蒙古工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP308

【参考文献】