A TCP Congestion Control Algorithm Based on Deep Reinforcement Learning Combined with Probe Bandwidth Mechanism

Abstract

The rapid development of emerging Internet services such as live video, 5G, VR, and the Internet of Things puts forward higher requirements for network throughput, Latency, jitter, and loss. However, the inefficient bandwidth utilization rate of the existing TCP protocol cannot meet these requirements. Based on this problem, this paper proposes an algorithm RL-explore that uses RL (Reinforcement learning) combined with bandwidth detection mechanism. The model trained with this algorithm can effectively use the network bandwidth, and compared to other RL algorithms, it is easier to converge during training.