A simulation study on the throughput fairness of TCP Vegas

Tbmugbput fairness is an important d e r i a for evaluating TCP pe fomance. Fairness is especial! important for best effort service, which is still the dominant gpe o f service in the Internet and, predictab!, in the years to come. However, the TCP pmtocols prevailing in the Internet, including TCP Tahoe and TCP Reno, are known to be unfair, especial! to connections with larger mund-tnp delays. Using the ns simulator, we have tbomugbb examined the fairness o f TCP Vegas focusing on three issues: I ) Is TCP Vegas real! fair to connections with largerpropagation delays? 2) what is the impact ofthe thresholds, a and p, used in TCP Vegas' congestion avoidance algorithm on fairness? and 3) when there is a mixture o f TCP Vegas and TCP Reno connections, are TCP Vegas and TCP Reno fair to each other? The simulation results support that TCP Vegas is still unfair to connections with larger propagation delays, for example, when a = I and = 3. But, unlike TCP Reno, the delay bias does not necessab4 increase as the del9 diflerence increases. The unfairnesspmblem can be resolved by an Enhanced TCP Vegas that sets a = p = 2 or 3 but not 1. When a = / . 3 = I , the@irness is unstable and may be worse than that when a = I and p = 3. Considering a trade-off among fairness, stabilig and agqressiveness, a value of3 seems to be an acceptab! good choice. Final!, fairness between TCP Reno and TCP Vegas connections depends on the RED gateway thresholds, the number ofactiveflows, and TCP Vegasparameters a and p.