A DQN-based CUBIC for TCP Congestion Control
Sang-Jin Seo, Geon-Hwan Kim, You-Ze Cho · 2022 27th Asia Pacific Conference on Communications (APCC) · 2022
In a static environment, as the available bandwidth increases, the slow congestion window increase rate prevents the existing TCP fully utilizing the bandwidth. CUBIC, a congestion control algorithm for high-speed networks, can provide higher throughput than existing algorithms but cannot guarantee satisfactory performance during frequent bandwidth fluctuations. Applying Deep-Q-Network to a TCP congestion control algorithm can improve link utilization. Therefore, in this paper, we propose a DQN-based CUBIC for various networks environments, which simultaneously utilizes the mechanisms of CUBIC and DQN. Through simulation experiments based on NS-3, it was confirmed that the proposed algorithm can increase the throughput compared to CUBIC in any link environment.