基于SDN的保证QoS的网络资源分配和管理

发布时间：2018-05-12 10:04

本文选题：SDN + QoS　；参考：《北京交通大学》2017年硕士论文

【摘要】：传统的QoS机制是为因特网而设计,建立在因特网完全分布式的、逐跳路由式的体系结构之上,缺乏整体网络资源分布的统一的全局视图,因而难以推广应用。SDN(Software Defined Network)具有集中控制的特点,通过集中控控制器能够轻松下发QoS策略,实现对所有网络设备以及全网流量的集中管理控制,既可以完成灵活的QoS服务策略选择,又能保证QoS策略的一致性。因而,在SDN下的QoS机制一经提出,就得到学术界和工业界的关注,成为研究热点。随着云计算、移动互联网、Web2.0等新兴业务的蓬勃发展,保证数据中心网络及其应用的服务质量早已成为业界共识。在数据中心网络中,许多著名的应用形式,例如Web搜索、广告、推荐系统等均具有类似的业务流特征,即是大流和小流混合的业务流。小流对时延敏感,但带宽要求不高;大流带宽要求高,但对时延不敏感。小流经历的时延直接影响到这些应用中返回结果的质量和最终的经济收益,因此如何保护小流的时延一直是研究热点。本论文第一部分工作提出了一种在SDN下保证小流时延的QoS机制。针对目前的网络节点输出端口的处理能力具有非抢占性的特点,即在一个时刻只能处理一个分组,且不能被抢占,该机制将分级队列管理和动态路由机制相结合,在输出端口没有被大流分组占用时,采用分级队列管理,保证小流分组优先服务;在输出端口被大流分组占用时,采用偏转路由方案,根据其他输出端口的占用情况,选择最优路由,保证小流分组优先传输。仿真实验表明采用该机制可以明显降低小流的传输时延。视频流媒体已成为目前因特网的主要应用,为了给视频用户提供更好的网络服务,了解并保证视频流的服务质量和用户体验成为网络提供商的重要任务。由于播放视频时发生卡顿会直接影响到用户的观看体验,从而影响用户观看视频的次数和时长,卡顿事件往往被用来作为衡量用户体验的主要指标。鉴于基于HTTP的动态自适应流媒体,即是当前视频流媒体的主要传输技术,本论文第二部分工作提出一种根据DASH视频流的QoS参数来预测视频卡顿的方法。该方法利用视频客户和服务器之间两个方向上的流量数据,使用机器学习技术,建立分类模型,预测视频播放时是否发生卡顿。原型实验结果显示,采用我们的方法预测实时卡顿事件的准确率高达98%,漏检率仅为3%。
[Abstract]:The traditional QoS mechanism is designed for the Internet, which is based on the distributed and hop-by-hop routing architecture of the Internet, and lacks a unified global view of the overall network resource distribution. Therefore, it is difficult to popularize the application of .SDNN Software Defined Network, which has the characteristics of centralized control. Through the centralized controller, the QoS policy can be easily sent out, and the centralized management control of all network devices and the whole network traffic can be realized. It can not only complete the flexible QoS service policy selection, but also guarantee the consistency of QoS policy. Therefore, once the QoS mechanism under SDN is put forward, it has attracted the attention of academia and industry, and has become a research hotspot. With the development of cloud computing, mobile Internet, Web 2.0 and other new businesses, it has become a consensus in the industry to ensure the quality of service of data center network and its applications. In data center networks, many well-known applications, such as Web search, advertising and recommendation systems, have similar traffic characteristics, that is, large and small streams of traffic. The small stream is sensitive to delay, but the bandwidth is not high, while the large stream is not sensitive to delay. The delay experienced by the stream directly affects the quality of the returned results and the final economic benefits in these applications. Therefore, how to protect the delay of the stream is always a hot topic. In the first part of this thesis, we propose a QoS mechanism to guarantee the stream delay in SDN. In view of the non-preemptive ability of the current network node output port, that is, only one packet can be processed at a time and can not be preempted, the mechanism combines hierarchical queue management with dynamic routing mechanism. When the output port is not occupied by the large stream packet, the hierarchical queue management is adopted to ensure the priority service of the small stream packet, and when the output port is occupied by the large stream packet, the deflection routing scheme is adopted, according to the occupation of other output ports, The best route is chosen to ensure the priority transmission of the small stream packet. Simulation results show that the proposed mechanism can significantly reduce the transmission delay of the stream. Video streaming media has become the main application of the Internet at present. In order to provide better network services for video users, understanding and guaranteeing the quality of service and user experience of video streams has become an important task for network providers. Because the occurrence of Catton when playing video will directly affect the user's viewing experience and thus affect the number and duration of the user watching the video, the Catton event is often used as the main index to measure the user experience. In view of the fact that dynamic adaptive streaming media based on HTTP is the main transmission technology of video streaming at present, the second part of this paper proposes a method to predict video Carton based on the QoS parameters of DASH video stream. This method uses the traffic data in two directions between the video client and the server, uses the machine learning technology, establishes the classification model, and predicts whether the video playback will happen or not. The prototype experiment results show that the accuracy of our method for predicting real time Catton events is as high as 98%, and the missed detection rate is only 3%.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【参考文献】