Research On Data Driven Mobile Video Live Streaming Congestion Control Methods

As the most critical algorithm in the network transmission part of the WebRTC standard protocol for network real-time communication,the Google congestion control algorithm GCC has been applied to many major browsers(including Chrome,Firefox,Opera,etc.).Although the goal of GCC is to achieve high throughput while maintaining low latency,we found that in mobile video live scenarios,GCC is too conservative in strong network(low latency and low packet loss).To be more specific,GCC regards the small jitter of network delay as a sign of network congestion,therefore often reduces the transmission throughput to avoid congestion.To address this issue,first,this paper obtains a data set of more than 1.18 million mobile Taobao live network traces from the partner Taobao(China)Software Co.,Ltd.,and verifies GCC's misjudgment problem under strong network.Secondly,this paper proposes a reconfigurable congestion control algorithm GCC-? based on GCC,which avoids unnecessary bit rate drop of up to 90%.In addition,this paper proposes a data-driven reinforecenment learning congestion control algorithm and verifies its performance.The prediction accuracy of the algorithm for network bandwidth is increased from 78.57%to 88.16%.In order to speed up verifying the performance of the congestion control algorithms,this paper designs and implements a live process simulator.With the GCC algorithm extracted from WebRTC,it takes only 10 minutes to perform a 50-hour live simulation.