流媒体播放任务调度与资源配置

发布时间：2018-05-15 04:34

本文选题：流媒体播放任务 + 调度　；参考：《清华大学》2016年博士论文

【摘要】：视频点播系统中的流媒体播放服务是近年来最热门的互联网服务之一。这类服务通常建立在大规模服务器集群上,以保障对大量用户需求的及时响应。在传统的自建型数据中心中,对流媒体播放任务的调度依据主要以数据局部性为主,并未考虑这类任务的时间周期变化性,使得集群服务器未得到最大程度的利用,造成设备成本、电力资源的浪费;在新兴的云平台虚拟集群中,单一种类的资源配置亦无法充分利用公共设备按需使用的特性,导致较高的租用成本。针对以上问题,本文提出基于节能的流媒体播放任务调度算法和基于成本控制的资源配置算法,以降低数据中心的使用成本,合理利用公共平台虚拟资源。主要工作从流媒体播放任务的特征分析与预测,调度与集群节能,以及云平台上的资源配置与成本控制三个方面展开。1.对视频点播系统中的流媒体播放任务特征进行分析,给出基于用户行为特征的任务长度预测方法,和基于历史数据特征的任务量预测方法,并对预测方法的准确性和开销进行了分析。对于流媒体播放任务的特征预测在对任务的调度和资源配置工作中起到决定性的作用,实验表明两种预测方法均有较高的准确性和较低的开销,能够有效地为后续算法进行服务。2.基于任务长度预测方法,提出了面向节能的流媒体播放任务在线调度算法,建立数学模型并利用真实数据完成模拟实验。本文发现长度特征对流媒体播放任务调度的影响,及传统调度方式对服务器集群电力资源造成的浪费;继而利用长度特征对流媒体播放任务进行分类,并根据任务种类通过排序、隔离等方法进行调度。实验表明该算法可大幅降低数据中心的电力开销,有效利用服务器集群。3.基于任务量预测方法,提出面向成本控制的流媒体播放任务资源配置算法,利用真实数据和商用云平台完成实验。该方法结合流媒体播放任务量周期性变化的特点,综合利用不同计价类型的云平台实例及存储资源,合理分配各类任务可使用的资源类型和资源的比例。实验表明该算法可在开销极低的情况下将云平台上各类资源开销降低至少20%。
[Abstract]:Streaming media service in VOD system is one of the most popular Internet services in recent years. Such services are usually built on large-scale server clusters to ensure timely response to a large number of user needs. In the traditional self-built data center, the scheduling of streaming media play task is mainly based on data locality, and the time cycle variation of this kind of task is not considered, so the cluster server is not utilized to the maximum extent. In the emerging cloud platform virtual cluster, the single type of resource allocation can not make full use of the characteristics of public equipment on demand, resulting in higher rental costs. In order to reduce the cost of using data center and make rational use of virtual resources of common platform, this paper proposes a scheduling algorithm of streaming media playback task based on energy saving and a resource allocation algorithm based on cost control. The main work is to analyze and predict the characteristics of streaming media play task, scheduling and cluster energy saving, and resource allocation and cost control on cloud platform. The task characteristics of streaming media play in VOD system are analyzed. The task length prediction method based on user behavior feature and the task quantity prediction method based on historical data feature are presented. The accuracy and cost of the prediction method are analyzed. The characteristic prediction of streaming media playback task plays a decisive role in task scheduling and resource allocation. Experiments show that the two prediction methods have higher accuracy and lower overhead. Can effectively service the subsequent algorithm. 2. Based on the method of task length prediction, an online scheduling algorithm for streaming media playback tasks for energy saving is proposed. The mathematical model is established and the simulation experiment is carried out using real data. This paper finds out the influence of length characteristics on the scheduling of streaming media playback tasks, and the waste of power resources caused by traditional scheduling methods, and then classifies the tasks of streaming media playing by using length characteristics. And according to the type of tasks through sorting, isolation and other methods to schedule. The experimental results show that the proposed algorithm can greatly reduce the power cost of the data center and make use of the server cluster. 3. 3. Based on the task quantity prediction method, a cost control oriented resource allocation algorithm for streaming media playback task is proposed, and the experiments are carried out using real data and commercial cloud platform. This method combines the characteristics of periodic variation of streaming media playback tasks, synthetically utilizes cloud platform examples and storage resources of different pricing types, and reasonably allocates the types of resources and the proportion of resources that can be used by all kinds of tasks. Experiments show that the algorithm can reduce the overhead of all kinds of resources on cloud platform by at least 20 parts.
【学位授予单位】：清华大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN948.64

【相似文献】