云环境下的集群性能优化研究

发布时间：2018-09-05 19:19

【摘要】：随着互联网的蓬勃发展,信息技术的广泛普及,云计算技术带给人们前所未有的网络服务体验。当前针对云计算的研究主要集中在云计算的模型架构、任务颗粒度任务调度策略容错机制和能耗优化等方面,如何对提交到云计算数据中心的任务集合进行合理调度及在任务执行时节约能耗,对该技术的发展具有非同寻常的意义,本文分别对任务调度策略和能耗优化两个方面进行了研究。云计算的虚拟化技术决定了其底层物理设备及数据中心的主机可以由异构化硬件环境组成,并且数据中心的规模有呈指数型增长的趋势,如何在大规模异构环境下对任务集合进行合理的资源分配是一个值得研究的课题,同时在执行任务的同时,如何使得所花费的能耗降到最低,也是一个研究热点。本文对云计算中的任务调度进行了研究,在分析云计算作业调度算法的需求的基础上,提出了调度算法在线性意义上的解空间,针对云计算的编程模型框架,提出了一种结合禁忌搜索的人工鱼群算法。该算法以任务总执行时间作为寻优函数,在网格化的解空间内运行,空间内的一个点代表一种解决方案,可以快速得出精确解,避免了基本人工鱼群算法不能获得精确解的缺陷。结合禁忌搜索思想,既保留了人工鱼群算法计算基数大仍能快速收敛的优点,又充分利用禁忌搜索不会陷入局部最优解的优势。其次,本文针对能耗优化的目标给出改进的AutoScale算法,该算法是在AutoScale算法基础上,引入最佳频率的思想,与AutoScale算法相比,该算法具有任务量与能耗比例最大化、在不增加额外能耗的情况下预留部分能力以应对突发负载的优点。最后针对上述的两种调度策略,设计了相应的仿真实验。实验结果表明,两度调度策略相比同类算法有着良好的性能表现。
[Abstract]:With the rapid development of the Internet and the wide spread of information technology, cloud computing technology brings people unprecedented experience of network services. The current research on cloud computing mainly focuses on the cloud computing model architecture, task granularity? Task scheduling policy? In the aspects of fault-tolerant mechanism and energy consumption optimization, how to reasonably schedule the task set submitted to the cloud computing data center and save energy consumption in task execution is of great significance to the development of this technology. In this paper, task scheduling strategy and energy consumption optimization are studied. The virtualization technology of cloud computing determines that the underlying physical devices and the host of the data center can be composed of isomerized hardware environment, and the size of the data center is increasing exponentially. How to allocate resources reasonably to task set in large-scale heterogeneous environment is a research topic. How to minimize the energy consumption while executing tasks is also a research hotspot. In this paper, the task scheduling in cloud computing is studied. On the basis of analyzing the demand of job scheduling algorithm in cloud computing, the solution space of scheduling algorithm in linear sense is proposed, and the programming model framework for cloud computing is proposed. An artificial fish swarm algorithm combined with Tabu search is proposed. The algorithm takes the total execution time of the task as the optimization function and runs in the grid solution space. A point in the space represents a solution, and the exact solution can be obtained quickly. The defect that the basic artificial fish swarm algorithm can not get the exact solution is avoided. In combination with the idea of Tabu search, the advantages of artificial fish swarm algorithm (AFS) in fast convergence are retained, and the advantage that Tabu search does not fall into local optimal solution is fully utilized. Secondly, this paper presents an improved AutoScale algorithm for the purpose of energy consumption optimization. This algorithm is based on the AutoScale algorithm and introduces the idea of optimal frequency. Compared with the AutoScale algorithm, the algorithm has the maximum ratio of task and energy consumption. The advantage of reserving a portion of the capacity to cope with a burst load without additional energy consumption. Finally, the corresponding simulation experiments are designed for the above two scheduling strategies. Experimental results show that the two-degree scheduling strategy has good performance compared with the same algorithm.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP3

【相似文献】