【作者】 李君；

【导师】 程春玲；

【作者基本信息】 南京邮电大学 ， 计算机应用技术， 2014， 硕士

【摘要】 随着云计算技术的普及和快速发展，到达云计算平台中的任务请求多种多样。为了满足不同种类的任务需求，组成云计算硬件平台的云数据中心中的计算节点不得不长期保持开启状态等待任务的到达，导致云数据中心对能耗的利用呈现出低利用、高浪费的特点。作为云计算的重要组成部分，任务调度是将任务映射到资源的过程，直接反应计算节点的使用情况及能耗开销。因此，通过合理的任务调度策略，可以实现云数据中心中的能耗优化。现有云数据中心的硬件资源通常是由大规模异构计算节点组成，且各计算节点之间通过具有不同传输率的链路链接而成。本文结合云计算的异构性，从任务的依赖性划分，以异构云计算平台中节能的独立任务调度和节能的依赖任务调度为研究目标，进行了三个方面的工作：（1）分析现有云数据中心的高能耗来源，阐述现有数据中心的能耗优化方法、国内外的研究现状，并从任务依赖性划分，重点对云计算平台中的节能的独立任务调度技术和节能的依赖任务调度技术进行分析。（2）到达云计算平台的独立任务具有随机性，使云计算中的计算节点不得不长期处于开启状态等待任务的到来而产生了能耗浪费。结合休假排队论的思想，提出一种节能的任务调度算法优化云计算中的能耗。首先，使用空竭服务的休假排队模型对异构云计算中的任务调度建模，其次，基于稳态条件下排队系统的忙期和忙循环对异构云计算中任务的平均响应时间与计算节点的平均功率进行分析，并提出一种基于相似任务的任务调度算法，以降低云计算平台中的能耗。仿真结果表明提出的算法能较好地保证任务性能，并有效降低云计算的能耗开销。（3）针对云计算平台中DAG任务图表示的依赖任务，提出了一种综合时间能耗成本的依赖任务调度算法。在确定任务调度次序的过程中，面对同一层中的任务只单独考虑时间因素过于单一的问题，引入计算因子和通信因子，综合时间与能耗成本共同确定任务调度的次序；在为任务选择计算节点阶段，结合计算节点的计算能力与执行能耗开销率及其局部任务队列共同确定计算节点的负载，以确定出未超过负载上限的计算节点集合，在该计算节点集合中，将任务分配到使其能最早开始的计算节点，并结合DVFS技术进行电压调节，仿真结果表明提出的算法具有很好的节能效果，实现了性能和能耗之间的平衡。更多还原

【Abstract】 With the popularity and rapid development of Cloud Computing technology, tasks coming toCloud Computing are infinite variety. In order to meet kinds of demands of tasks, computing nodesin the Cloud Computing platform have to remain open to wait for tasks, which makes energyefficiency of cloud data center has characteristic of low utilization, high waste. As an important partof Cloud Computing platform, task scheduling is the process of tasks directly mapped to resourcesof cloud data center, which directly responds energy cost of data center resources. Therefore, energyoptimization of cloud data center can be implemented by reasonable task scheduling strategy. Clouddata center is usually composed of large-scale heterogeneous computing nodes which are linked bydifferent transmission rates. From the dependence of the division of tasks, this thesis takesenergy-saving independent task scheduling technology and energy-saving dependent taskscheduling technology in heterogeneous cloud computing platform as research objectives. Themajor work in this thesis can be included from three aspects:(1) The existing sources of high-energy consumption of the cloud data center are analyzed, theenergy optimization methods in existing data centers, research status are expounded, the problem ofindependent task scheduling for energy-saving and dependent task scheduling for energy-saving ofheterogeneous Cloud Computing platform are analyzed.(2) Tasks coming to the Cloud Computing are random and computing nodes of CloudComputing have to remain open to wait for tasks, which produces much wasted energy. Anenergy-saving task scheduling algorithm based on vocation queuing model for the CloudComputing is presented. First, task scheduling model of the heterogeneous Cloud Computing isestablished by using exhaustive service, vacation queuing model system. And then, the averageresponse time of tasks and the average power of heterogeneous compute nodes are analyzed byusing the busy period and busy spin under steady state conditions in the Cloud Computing. Afterthat, a task scheduling algorithm based on similar tasks is proposed to reduce energy consumption.Simulation results show that the proposed algorithm can ensure the task performance, and reducethe energy consumption of the Cloud Computing effectively.(3) For the dependent task presented as DAG graph of the cloud computing platform, thisthesis proposes a task scheduling algorithm based on time and energy consumption cost inheterogeneous cloud. During the process of determining scheduling order of tasks, computing factor and communication factor are introduced by integrating time and energy costs to determine theorder of task scheduling; in the phase of selecting the computing nodes for task, computing node’sload is redefined by combining with its ability to calculate task, rate of executing energyconsumption and the amount of tasks in the local task queue. Then the set of computing nodes thatdo not exceed the threshold value of load limit is determined. Task is assigned to the computingnode which can first begin to perform a task in the set of computing nodes and DVFS technique iscombined for voltage regulation. Simulation results show the good energy-saving effect of theproposed algorithm.更多还原