This article reviews the current transmission control protocol (TCP) congestion control mechanisms and recent advances in analytical tools. It introduces an optimization-based framework that interprets various flow control mechanisms, particularly the utility being optimized by the protocol's equilibrium structure. The dynamics of TCP are analyzed using linear models, revealing stability limitations despite built-in compensations for delay. A new protocol is proposed that overcomes these limitations and provides stability scalable to arbitrary networks, link capacities, and delays. The article begins by highlighting the importance of understanding congestion control in the expanding and diverse Internet. It discusses the limitations of heuristic and intricate control mechanisms and the recent progress in analytical modeling. Key to these advances is the explicit modeling of congestion measures, such as loss probability or queueing delay, which are used to communicate congestion information to data sources. Two types of studies are emphasized: characterizing equilibrium conditions from a congestion control protocol's perspective and analyzing the dynamics of these protocols, particularly their stability in the presence of feedback delay. Recent analysis has shown that predominant TCP implementations, such as Reno, are prone to instabilities when combined with network delays and increased capacity. The article then describes the current TCP protocols, focusing on window flow control and the Tahoe and Reno algorithms. It explains how TCP infers congestion and adjusts window size, with Reno refining the initial Tahoe algorithm to improve loss recovery efficiency. The analytical models developed in recent years are discussed, including the use of deterministic flow models and the assumption that sources have access to the aggregate price of all links in their route. The equilibrium structure and utility optimization are explored, showing how sources maximize individual profit based on their utility functions. The dynamics and stability of TCP Reno/RED are analyzed, revealing that the equilibrium is often unstable and leading to oscillations. The article also discusses the impact of delays and buffer sizes on loss rates and queueing delays, and the fairness issues in TCP Reno. Finally, the article presents a new protocol that overcomes the stability limitations of existing TCP variants, providing linear stability for arbitrary delays, capacities, and routes.This article reviews the current transmission control protocol (TCP) congestion control mechanisms and recent advances in analytical tools. It introduces an optimization-based framework that interprets various flow control mechanisms, particularly the utility being optimized by the protocol's equilibrium structure. The dynamics of TCP are analyzed using linear models, revealing stability limitations despite built-in compensations for delay. A new protocol is proposed that overcomes these limitations and provides stability scalable to arbitrary networks, link capacities, and delays. The article begins by highlighting the importance of understanding congestion control in the expanding and diverse Internet. It discusses the limitations of heuristic and intricate control mechanisms and the recent progress in analytical modeling. Key to these advances is the explicit modeling of congestion measures, such as loss probability or queueing delay, which are used to communicate congestion information to data sources. Two types of studies are emphasized: characterizing equilibrium conditions from a congestion control protocol's perspective and analyzing the dynamics of these protocols, particularly their stability in the presence of feedback delay. Recent analysis has shown that predominant TCP implementations, such as Reno, are prone to instabilities when combined with network delays and increased capacity. The article then describes the current TCP protocols, focusing on window flow control and the Tahoe and Reno algorithms. It explains how TCP infers congestion and adjusts window size, with Reno refining the initial Tahoe algorithm to improve loss recovery efficiency. The analytical models developed in recent years are discussed, including the use of deterministic flow models and the assumption that sources have access to the aggregate price of all links in their route. The equilibrium structure and utility optimization are explored, showing how sources maximize individual profit based on their utility functions. The dynamics and stability of TCP Reno/RED are analyzed, revealing that the equilibrium is often unstable and leading to oscillations. The article also discusses the impact of delays and buffer sizes on loss rates and queueing delays, and the fairness issues in TCP Reno. Finally, the article presents a new protocol that overcomes the stability limitations of existing TCP variants, providing linear stability for arbitrary delays, capacities, and routes.