To allocate the bandwidth for internet traffic, we consider a mathematical model of inter-net traffic and compute the effective bandwidth based on given quality of service parameter. For doing this, we first analyze the stochastic characteristics of internet traffic trace obtained from a major backbone link. From our analysis, it can be shown that a nonhomogeneous Poisson process and a Gaussian process are good candidates to model internet traffic mathematically. To compute the effective bandwidth, we use a Gaussian process simply because a nonhomogeneous Poisson process has a number of parameters to describe. Our simulation studies based on the NS-2 simulator show that a bufferless model does not predict the effective bandwidth well. To solve this problem, we consider a buffered model with a special process of a self-similar Gaussian process, called the fractional Brownian motion(fBm). Our simulation studies show that the effective bandwidth from the buffered model based on a fBm fits well for internet traffic.