A Network Packet Loss and Congestion Control Method using Utility Function Objective Optimization Model

Yun Ju, Yunhao Hu

Abstract: To solve the congestion control algorithm of a dynamic network difficult to determine the appropriate size of the congestion window problem. To improve the traditional congestion control algorithm of the UDP black-box model, the packet loss behavior that is not congested or caused by congestion is distinguished. Optimizing the traditional PCC black-box model based on utility function, and improved PCC-DRL optimization algorithm based on the PCC method were proposed. Compared with the existing mainstream congestion control algorithm, the comparison results show that the application of THE PCC-DRL optimization algorithm improves the dynamic network bandwidth utilization rate 9.67%, reduces the packet loss rate 0.24%, reduces the delay 5.69ms, and improves the queue concurrency 6.73%. These results indicate that THE PCC-DRL algorithm has a good effect on distinguishing the packet loss behaviors caused by non-congestion or congestion in dynamic networks, and has good adaptability and robustness to dynamic conditions and congestion forms.

Keywords: Utility functions, Deep reinforcement learning, PCC, Congestion control