FAST TCP is a new congestion control algorithm designed for high-speed, long-latency networks. It addresses the four main difficulties of the current TCP implementation—Reno—when large bandwidth-delay products are involved: slow packet-level increase, drastic flow-level decrease, oscillation in congestion windows, and unstable dynamics. FAST TCP uses queueing delay as a congestion measure, which is more accurate and easier to estimate than loss probability. This approach helps maintain stability as the network scales up. The architecture of FAST TCP is divided into four functionally independent components: data control, window control, burstiness control, and estimation. The window control component, which is the focus of this paper, updates the congestion window based on the average RTT and a protocol parameter that determines the total number of packets queued in routers at equilibrium. The algorithm is analyzed mathematically to show that it achieves the same equilibrium properties as TCP Vegas, including weighted proportional fairness and stability. Experimental results demonstrate that FAST TCP outperforms other TCP variants in terms of throughput, fairness, stability, and responsiveness. It converges rapidly to equilibrium rates, maintains low queueing delays, and exhibits less oscillation compared to loss-based protocols like Reno, HSTCP, and STCP. simulations in ns-2 further validate the performance of FAST TCP in complex environments, including the interaction with Reno and multi-link networks. Open issues include accurate measurement of propagation and queueing delays, handling of heterogeneous protocols, and designing practical distributed algorithms for achieving desired inter-protocol fairness.FAST TCP is a new congestion control algorithm designed for high-speed, long-latency networks. It addresses the four main difficulties of the current TCP implementation—Reno—when large bandwidth-delay products are involved: slow packet-level increase, drastic flow-level decrease, oscillation in congestion windows, and unstable dynamics. FAST TCP uses queueing delay as a congestion measure, which is more accurate and easier to estimate than loss probability. This approach helps maintain stability as the network scales up. The architecture of FAST TCP is divided into four functionally independent components: data control, window control, burstiness control, and estimation. The window control component, which is the focus of this paper, updates the congestion window based on the average RTT and a protocol parameter that determines the total number of packets queued in routers at equilibrium. The algorithm is analyzed mathematically to show that it achieves the same equilibrium properties as TCP Vegas, including weighted proportional fairness and stability. Experimental results demonstrate that FAST TCP outperforms other TCP variants in terms of throughput, fairness, stability, and responsiveness. It converges rapidly to equilibrium rates, maintains low queueing delays, and exhibits less oscillation compared to loss-based protocols like Reno, HSTCP, and STCP. simulations in ns-2 further validate the performance of FAST TCP in complex environments, including the interaction with Reno and multi-link networks. Open issues include accurate measurement of propagation and queueing delays, handling of heterogeneous protocols, and designing practical distributed algorithms for achieving desired inter-protocol fairness.