Congestion Control Design of Multicast QUIC Based on Reinforcement Learning

Quick UDP Internet Connections (QUIC), an Internet protocol proposed by Google to replace TCP, is already used on YouTube and Facebook. With the popularization of new multimedia applications such as VR and the development of low-cost supporting equipment, the current QUIC protocol needs to further evolve to support new multimedia data and multicast applications. Therefore, it is a promising research direction to study how to apply QUIC protocol to multicast scenario and design congestion control algorithm in multicast scenario. The working mechanism of traditional network congestion control protocols is relatively simple, and they do not have the ability to adapt to different network environments. Based on the multicast QUIC protocol, this paper proposes a reinforcement learning congestion control algorithm based on Decaying epsilon-greedy and critic-actor methods. The simulation results show that the proposed algorithm can greatly reduce the packet loss rate and has good adaptability in the complex and changeable network environment.