This paper presents an analytical model for the TCP Congestion Avoidance algorithm, which predicts the bandwidth of a sustained TCP connection under light to moderate packet losses. The model assumes that TCP avoids retransmission timeouts and has sufficient receiver window and sender data. It is verified through simulations and live Internet measurements, comparing the model's predictions with actual TCP implementations and TReno diagnostic results. The model is also applied to analyze bandwidth allocation in networks with multiple congested gateways and to understand the behavior of the Internet under high load from diverse user communities. The paper discusses the limitations of the model, such as its inability to predict performance under conditions where pure Congestion Avoidance does not fully control TCP performance, and highlights the importance of appropriate measurements for parameters like packet loss rate and round-trip time.This paper presents an analytical model for the TCP Congestion Avoidance algorithm, which predicts the bandwidth of a sustained TCP connection under light to moderate packet losses. The model assumes that TCP avoids retransmission timeouts and has sufficient receiver window and sender data. It is verified through simulations and live Internet measurements, comparing the model's predictions with actual TCP implementations and TReno diagnostic results. The model is also applied to analyze bandwidth allocation in networks with multiple congested gateways and to understand the behavior of the Internet under high load from diverse user communities. The paper discusses the limitations of the model, such as its inability to predict performance under conditions where pure Congestion Avoidance does not fully control TCP performance, and highlights the importance of appropriate measurements for parameters like packet loss rate and round-trip time.