【作者】 杨凯；

【导师】 张旭； 马展； 曹汛；

【作者基本信息】 南京大学 ， 电子与通信工程（专业学位）， 2021， 硕士

【摘要】 随着网络基础设施建设以及无线通信技术的发展,视频传输在IP网络流量中占比超过80%。基于视频传输的新型应用层出不穷,可留言视频点播、可交互视频直播以及基于实时视频传输的云游戏等新型网络应用给用户带来更好的观看体验,受限于现有网络传输能力,例如云游戏这样典型的需求:更高清、更实时的视频传输体验还并不让人满意,如何在现有网络服务能力下提高视频传输质量成为越来越重要的研究方向。本文首先从云游戏的前身:网络游戏系统测量角度对相关文献进行总结和归纳,内容包括网络游戏发展历程,现有文献中网络游戏系统测量中所使用到的一些普适的或者特殊的测量方法,以及在测量过程中研究者最为关心的测量对象等等,这些关于网络系统测量的全面了解为接下来进行的实际测量提供了理论和方法上的指导,也对云游戏视频传输系统的仿真建模也提供系统原型的参考。利用总结得到的系统测量的相关经验,本文选择快速发展的国内视频直播服务系统,考虑到面向国内视频直播平台的系统测量工作还较为缺乏,本文首次对国内视频直播平台进行多方位的、全面深入测量分析,并与国外相同类型平台对比,着眼于视频直播平台的系统架构以及用户行为分析,结合新型可交互视频直播的特点,发现观众偏好和消费行为模式。针对测量过程中发现现有系统存在的一些问题中最典型的资源分配不公平,结合未来视频传输的热点云游戏视频传输方向,本文对实时多用户动态码率视频传输系统进行建模,不同于现有工作基于视频块的仿真而是基于视频帧级别的仿真工作,并同时解决了多用户网络传输仿真问题的建模工作,在此基础上应用多智能体强化学习的办法验证了多智能体强化学习在多用户视频传输问题中的实验效果,证明该仿真系统的实用性与可拓展性。综上所述,面对快速增长的视频传输实时性的需求与缓慢发展的网络服务能力之间的矛盾,经过充分的文献调研与分析,通过对现有新型视频直播平台传输服务的系统测量,揭示了其服务能力与用户特点,并依据需求冲突最明显的云游戏场景进行传输建模工作,以及对多用户实时视频传输问题进行仿真建模并提出可行的解决办法,本文的研究成果对了解视频传输系统特性,提升服务质量、降低服务成本具有一定的学术意义和参考价值。更多还原

【Abstract】 With the development of network infrastructure and wireless communication technology,video transmission accounts for more than 80% of IP network traffic.New applications based on video transmission are emerging in endlessly.New network applications such as video-on-demand messages,interactive video live broadcasts,and cloud games based on real-time video transmission will bring users a better viewing experience and are limited by existing network transmission capabilities,such as the typical demand for cloud gaming: a more high-definition,more real-time video transmission experience is not satisfactory.How to improve the quality of video transmission under the existing network service capabilities has become an increasingly important research direction.This thesis first summarizes the relevant literature from the measurement perspective of online game systems,the predecessor of online games.The content includes the development history of online games,and some general or special measurement methods used in the measurement of online game systems in the existing literature.As well as the measurement objects that researchers are most concerned about during the measurement process,etc.These comprehensive understandings of network system measurement provide theoretical and methodological guidance for the next actual measurement,and also provide a simulation model for the cloud game video transmission system.The model also provides a reference for the system prototype.Based on the relevant experience of system measurement,this thesis chooses the fast-developing domestic video live broadcast service system.Considering that the system measurement work for domestic video live broadcast platforms is still lacking,this thesis is the first time to conduct a multi-faceted and comprehensive in-depth study of domestic video live broadcast platforms.Measurement and analysis,and comparison with the same type of foreign platforms,focusing on the system architecture of the live video platform and user behavior analysis,combined with the characteristics of the new interactive live video,discovering audience preferences and consumption behavior patterns.Aiming at the most typical unfair resource allocation among the problems found in the existing system during the measurement process,combined with the hot cloud game video transmission direction of future video transmission,this thesis models a real-time multi-user dynamic bit rate video transmission system,which is different from Existing work is based on video block simulation but based on video frame level simulation work,and at the same time solves the modeling work of the multi-user network transmission simulation problem.On this basis,the multi-agent reinforcement learning method is used to verify the multi-agent The experimental effect of reinforcement learning in the problem of multi-user video transmission proves the practicability and scalability of the simulation system.To sum up,there is a contradiction between the fast-growing real-time demand for video transmission and the slow-developing network service capabilities.After thorough literature research and analysis,through the systematic measurement of existing new video transmission services,service capabilities and user characteristics are revealed.Based on the cloud gaming scenario with the most obvious demand conflict,this thesis carries out transmission modeling work,simulates and models multi-user real-time video transmission problems and proposes feasible solutions.The research results of this thesis are important for understanding video transmission systems.Features,improving service quality and reducing service costs have certain academic significance and reference value.更多还原