云计算环境下的网络服务器集群节能技术研究和绿色配置策略设计

发布时间：2018-03-28 23:22

本文选题：云计算　切入点：数据中心　出处：《合肥工业大学》2014年硕士论文

【摘要】：随着社会对云计算需求的持续扩大,规模巨大的数据中心不断被建立,而维护其运行需要大量的能量。随着传统能源的逐渐枯竭和价格的飞涨,维护运行数据中心的费用将会超过购买系统硬件的费用。因此如何高能效地运行数据中心是一个亟待解决的问题。近年来的研究热点是基于“绿色”配置理念,即使得数据中心的能量消耗与系统负载成正比。然而,当前的云计算数据中心(比如HOTMAIL和微软研究院)也远达不到这个目标,甚至系统空闲时的能量消耗也达到其峰值的一半左右。由于面向云计算的数据中心一般由大量服务器组成,一种广泛被国内外学者接受的有效绿色配置方式,是使用软件动态的调整活跃服务器数量与系统负载相匹配；在进行系统任务调度时进行合理的分配负载使得某些服务器,在较低负载时间段内无需承担任何任务进程,从而进入低功耗模式以节约能源。本文的研究对象为同构服务器组成的数据中心,将其服务运行过程划分为若干时间周期,根据每个周期内的负载信息基于历史统计或在线预测获得,将“绿色配置”策略分为离线算法和在线算法。离线算法适用于每个周期负载已知的情况,其最优化结果一方面提供活跃服务器数量精准的评估,另一方面为在线算法设置了较好的预测基础。首先在任意周期内考虑操作能耗以及在相邻周期之间考虑切换能耗,建立了数据中心能耗最小化数学模型；针对模型中出现的非线性约束,通过公式重建的方法进行简化。作为后续讨论最优离线算法的基础,首先对两种特殊情况(无操作能耗和无切换能耗)下的最优解进行讨论,进一步引出一般情况下的最优解特性,并给出了相关引理及证明过程。其次,给出了服务器绿色配置的最优离线算法。算法基于整数动态规划思想,一般具有指数复杂度。通过消除动态规划中的递推过程,其具有多项式复杂度。针对于不同负载变化趋势的数值结果表明,所提最优离线算法能保证能耗达到最小的同时计算过程平稳。最后,针对工作负载可能遇到的最差情况,设计了在线算法,该算法沿用了滚动时域控制的思想。通过仿真与最优离线算法的结果进行比较,得出在线算法为近似最优的结论。保持合适数量的活跃服务器提供及时服务的同时,尽可能地切换服务器到空闲模式以减少能耗是一种有效的节能措施。本文的工作为云计算数据中心的绿色配置技术提供了一定程度的理论成果。
[Abstract]:As society continues to grow in demand for cloud computing, huge data centers are being built, and maintaining them requires a lot of energy. The cost of maintaining and running the data center will exceed the cost of purchasing the hardware of the system. Therefore, how to run the data center efficiently is an urgent problem. In recent years, the research focus is based on the concept of "green" configuration. Even if the energy consumption of the data center is proportional to the system load, today's cloud computing data centers, such as HOTMAIL and Microsoft Research, are far from achieving this goal. Even when the system is idle, the energy consumption is about half of its peak. Because the data center for cloud computing is generally composed of a large number of servers, it is an effective green configuration method widely accepted by domestic and foreign scholars. Is to use software to dynamically adjust the number of active servers to match the system load; to distribute the load so that some servers do not have to undertake any task processes during lower load periods, In order to save energy, the research object of this paper is the data center composed of isomorphic servers, and the service running process is divided into several time periods. According to the load information in each cycle based on historical statistics or online prediction, the "green collocation" strategy is divided into offline algorithm and online algorithm. The optimization results provide an accurate evaluation of the number of active servers on the one hand, and a better prediction basis for online algorithms on the other. First, the operation energy consumption is considered within any period and the switching energy consumption is considered between adjacent cycles. The mathematical model of data center energy consumption minimization is established, and the nonlinear constraints in the model are simplified by the method of formula reconstruction, which is the basis of the following discussion on the optimal off-line algorithm. First, the optimal solutions of two special cases (no operating energy consumption and no switching energy consumption) are discussed, and then the properties of the optimal solutions in general cases are derived, and the relevant Lemma and the proof process are given. The optimal off-line algorithm of server green configuration is presented. The algorithm is based on integer dynamic programming and has exponential complexity. It has polynomial complexity. The numerical results for different load trends show that the proposed optimal off-line algorithm can ensure the minimum energy consumption and smooth calculation process. Finally, for the worst case that the workload may encounter, An online algorithm is designed, which adopts the idea of rolling time domain control. The simulation results are compared with the results of the optimal off-line algorithm. Draw the conclusion that the online algorithm is approximately optimal. While maintaining a suitable number of active servers to provide timely services, Switching servers to idle mode as much as possible to reduce energy consumption is an effective energy saving measure. The work of this paper provides a certain degree of theoretical results for the green configuration technology of cloud computing data center.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【相似文献】