Methods of Congestion Control in Wired Networks with Reinforcement Learning – A Review

In a TCP/IP network, the TCP congestion control system (CC) ensures that network resources are shared efficiently and fairly between its users. Previously, TCP CC systems have been designed to cable hard from predefined behaviour to particular network control signal. However, as networks become more complex and competitive, the optimal feedback action mapping the invention of this analysis will be difficult to develop, and the contribution of this study is the thorough description of and comparison of the congestion control (CC) approaches. Comparatively to standard CC algorithms, which are primarily rule-based, the capacity to learn from previous knowledge is extremely valuable. According to the research, RL is a major trend among learning-based CC algorithms. In the paper we discuss the efficiency of CC algorithms on an RL basis and present current issues with CC algorithms on an RL basis. We describe the problems and dynamics of CC algorithms based on learning.