On the Gaussian Characteristics of Aggregated Short-Lived Flows on High-Bandwidth Links

Abstract

Traffic modeling, traffic decomposition, and traffic engineering are some of the applications of traffic characterization that are mainly based on statistical characteristics of the network traffic. Many empirical analyses on Internet traffic traces show that the flow inter-arrival time distribution generally follows the Weibull distribution. As the scale of the network becomes larger, the Weibull distribution degrades to the Poisson distribution and when the flow arrival rate is high, it asymptotically converges to the Normal distribution. The aggregated traffic on high bandwidth links is the result of statistical multiplexing of many traffic sources, and the flow arrival rate on these links is sufficiently large. In this paper, using empirical analysis conducted by means of our Trace Analyzing Tool, along with ns2 simulations, we show that the aggregated short-lived flows on high-bandwidth links show Gaussian characteristics. Hence, using an adequate denoising filter, the short-lived flows can be separated.