Um estudo comparativo dos algoritmos TCP Reno e TCP Vegas sobre uma rede de serviços diferenciados, em termos de distribuição de banda passante

Abstract

It is presented a study, based on simulations, about the influence of α and β parameters included in TCP Vegas algorithm as well as the influence of the RIO algorithm parameters, on fairness of a Differentiated Services (DiffServ) network. Fairness is treated here as the capacity of the network in sharing its bandwidth proportionally to the Service requested for each user. The TCP Reno algorithm is simulated in a way that allows its results to be compared with the ones from TCP Vegas. It is also evaluated the interoperation between these two TCP algorithms and the effíciency of each one with regards to the UDP aggressiveness. Two simulation models are employed, the fírst one with only two sources and the other one with ten sources. Three sets of values for the RIO algorithm and four sets of values for α and β parameters are evaluated too. The analysis of the results revealed that the unfairness problem demonstrated in recent researches related to the TCP Vegas in conventional IP networks, also happens in a Diffôerv environment, and that the adequate adjustment for the mentioned parameters can minimize such problems. The best results were obtained from the model which consists of two sources, by using α and β and “extremely small” values for RIO. In this case, a good interoperation was gotten between both TCP algorithms Regarding to the ten-sources model, no important advantage was observed when adjusting those parameters and in most cases TCP Reno surpassed TCP Vegas. The UDP traffic showed an aggressive behavior against both TCP algorithms, regardless of the employed adjustments.