Congestion Control in IP Networks using Machine Intelligence

Congestion control (CC) across the entire network serves as a crucial component of TCP, effectively managing packet loss caused by network congestion resulting from overload. The emergence of numerous network technologies such as SD-WAN, Wi-Fi 6, 5G cellular, network automation, and satellite communications has introduced a multitude of diverse network data transmission scenarios. This expanded landscape poses significant challenges to conventional CC methods, which rely on rule-based approaches and may not guarantee optimal performance across diverse network environments. Consequently, there is a demand for an intelligent CC module that facilitates seamless communication while maximizing resource utilization. Machine learning techniques have demonstrated their significant role in various applications, including communication, networking, and network management. Notably, learning-based TCP CC schemes have garnered considerable attention due to their ability to learn from network interactions and adapt accordingly. This research endeavour aims to develop an experience-driven approach that empowers the congestion module to learn optimal congestion control strategies based on its own historical data, such as runtime statistics. By leveraging reinforcement machine learning algorithms, the proposed approach enables dynamic and intelligent decision-making, ultimately minimizing packet loss and enhancing network performance.