TCP Vegas: New Techniques for Congestion Detection and Avoidance

TCP Vegas: New Techniques for Congestion Detection and Avoidance

1994 | Lawrence S. Brakmo, Sean W. O'Malley, Larry L. Peterson
This paper introduces TCP Vegas, a new implementation of TCP that significantly improves throughput and reduces packet losses compared to the Reno implementation. The key techniques employed by Vegas include a more accurate retransmission mechanism, a proactive congestion avoidance mechanism, and a modified slow-start mechanism. The retransmission mechanism uses system timestamps to detect and retransmit lost segments more promptly. The congestion avoidance mechanism adjusts the congestion window based on the difference between expected and actual throughput rates, aiming to maintain an optimal amount of extra data in the network. The modified slow-start mechanism allows for exponential growth only every other round-trip time, reducing the risk of congestion during the initial phase of a connection. Experimental results from both simulations and real-world measurements on the Internet show that Vegas achieves 40 to 70% better throughput with significantly fewer retransmissions compared to Reno. The paper also discusses the limitations and future improvements, emphasizing the need for further testing and congestion avoidance strategies during slow-start.This paper introduces TCP Vegas, a new implementation of TCP that significantly improves throughput and reduces packet losses compared to the Reno implementation. The key techniques employed by Vegas include a more accurate retransmission mechanism, a proactive congestion avoidance mechanism, and a modified slow-start mechanism. The retransmission mechanism uses system timestamps to detect and retransmit lost segments more promptly. The congestion avoidance mechanism adjusts the congestion window based on the difference between expected and actual throughput rates, aiming to maintain an optimal amount of extra data in the network. The modified slow-start mechanism allows for exponential growth only every other round-trip time, reducing the risk of congestion during the initial phase of a connection. Experimental results from both simulations and real-world measurements on the Internet show that Vegas achieves 40 to 70% better throughput with significantly fewer retransmissions compared to Reno. The paper also discusses the limitations and future improvements, emphasizing the need for further testing and congestion avoidance strategies during slow-start.
Reach us at info@study.space