数据驱动的移动视频直播拥塞控制技术研究
首发时间:2019-04-22
摘要:谷歌拥塞控制算法(GCC)作为网络实时通信(WebRTC)标准协议中网络传输部分最关键的算法,已经应用到主流浏览器(包括Chrome和Firefox等)中。虽然GCC的目标在与实现高吞吐量的同时保持低延迟,但我们发现GCC的性能远不能令人满意,尤其是在良好的网络条件下。因此,本课题主要研究基于WebRTC框架的拥塞控制算法的优化。为此,本课题从国内主流直播平台获取了超过118万个手机直播会话的网络踪迹(Trace)数据集,设计并实现了一个数据驱动的直播过程模拟器,消除了拥塞控制算法等待视频帧编解码和路由传输的时间,配合从WebRTC中提取的GCC算法,进行50小时的模拟直播只需要10分钟。此外,本课题提出了一个基于强化学习的拥塞控制算法,算法对于网络带宽的预测准确率从78.57%提升到88.16%。
For information in English, please click here
Research on Data Driven Mobile Video Live Streaming Congestion Control Methods
Abstract:Google congestion control (GCC) is the most significant algorithm in the networking module of web real-time communicating standard WebRTC, and has been implemented in mainstream browsers including Chrome and Firefox. While GCC\'s aiming at achieving high video sending rate and low latency simultaneously, we find that GCC\'s performance is far from satisfactory particularly under good network conditions. Hence in this paper we focus the improvement of congestion controlling under WebRTC.In particular, we collect a GCC trace dataset with over 1.18 million sessions from a main stream live media streaming platform and design and implement a data driven live streaming simulator, which eliminates the codec and routing time. With the GCC extracted from WebRTC, it needs only 10 minutes to simulate 50 hours of live streaming. In addition, we propose a reinforcement learning based congestion control algorithm, improving accuracy of bandwidth prediction from 78.57% to 88.16%.
Keywords: low-latency congestion control real-time video transmission reinforcement learning
基金:
引用
No.****
同行评议
共计0人参与
勘误表
数据驱动的移动视频直播拥塞控制技术研究
评论
全部评论0/1000