基于Docker的视频监控云平台资源预测与配置研究

发布时间：2018-07-07 12:07

本文选题：云计算 + Docker　；参考：《北京邮电大学》2017年硕士论文

【摘要】：随着视频监控设备的大规模部署,传统的视频监控系统已无法对海量的视频监控数据进行有效的计算分析,伴随云计算的发展,视频监控系统与云计算融合所形成的视频监控云成了近年来的研究热点。当前主流的视频监控云平台基本都是以虚拟机的形式为用户提供服务,资源分配粒度大且存在性能损耗。目前云平台中对资源配置基本是在初始阶段采取静态分配方式,然后在运行期再根据监控预警或负载预测的方法进行动态水平伸缩。然而监控预警的方式没有考虑资源需求的实时性,是一种被动伸缩方式,容易违反服务等级协议SLA;负载预测方式虽然具有预动性,但目前仍没有一种预测方法能够对视频监控云平台中的服务负载做出准确的预测。如何提高视频监控云平台的资源利用率,实现一个高效的可以动态弹性伸缩的视频监控云平台是本文的研究重点。首先,本文通过对典型视频监控服务的负载特性进行分析,提出了适用于视频监控云平台的资源预测模型,该模型的构建可分为两个阶段,在第一阶段可以根据视频服务的属性特征来预测其初始资源需求量,第二阶段基于资源需求时间序列相似性对工作负载进行预测。其次,针对视频监控云平台中无法对资源及时准确的进行动态调整这一问题,本文对现有基于容器技术的视频监控云平台进行了优化,在云平台中添加了资源预测模块并调整了资源配置策略,使之可以根据预测结果通过热迁移及垂直伸缩的方式对容器的资源进行重配置,从而提高资源利用率。最后,本文实现了基于Docker的视频监控云计算平台的优化工作,对所提出的资源预测模型及调整后的资源配置策略进行了性能测试,实验结果表明本文所提出的资源预测模型有更高的准确率,所调整资源配置策略可以有效提高视频监控云平台的资源利用率。
[Abstract]:With the large-scale deployment of video surveillance equipment, the traditional video surveillance system has been unable to calculate and analyze the mass of video surveillance data effectively, with the development of cloud computing. Video surveillance cloud formed by the fusion of video surveillance system and cloud computing has become a research hotspot in recent years. At present the mainstream video surveillance cloud platform is basically in the form of virtual machine to provide services to users resource allocation granularity and performance loss. At present, resource allocation in the cloud platform is based on static allocation in the initial stage, and then dynamically scalable in the runtime according to the method of monitoring, warning or load forecasting. However, the method of monitoring and warning does not take into account the real-time requirements of resources, it is a passive expansion mode, and it is easy to violate the service level protocol slaa. However, there is still no prediction method to accurately predict the service load in video surveillance cloud platform. How to improve the resource utilization of video surveillance cloud platform and realize an efficient and dynamic elastic video monitoring cloud platform is the focus of this paper. Firstly, by analyzing the load characteristics of typical video surveillance services, this paper proposes a resource prediction model for video surveillance cloud platform, which can be divided into two stages. In the first stage, the initial resource demand can be predicted based on the attribute characteristics of the video service. In the second stage, the workload is predicted based on the similarity of the time series of resource requirements. Secondly, aiming at the problem that the resources can not be adjusted dynamically and accurately in the video surveillance cloud platform, this paper optimizes the existing video surveillance cloud platform based on container technology. The resource prediction module is added to the cloud platform and the resource allocation strategy is adjusted so that it can reconfigure the container resources by heat transfer and vertical expansion according to the prediction results so as to improve the resource utilization ratio. Finally, this paper realizes the optimization of the video surveillance cloud computing platform based on Docker, and tests the performance of the proposed resource prediction model and the adjusted resource allocation strategy. The experimental results show that the proposed resource prediction model has higher accuracy and the resource allocation strategy can effectively improve the resource utilization of video surveillance cloud platform.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09

【参考文献】