Traffic Models in Broadband Networks

0163-6804/97/$10.00 © 1997 IEEE he need for telecommunication networks capable of providing diverse and emerging communication services such as data, voice, and video, motivated the standardization of broadband networks. The need for a flexible design that can accommodate future services and advances in technology led International Consultative Committee for Telephone and Telegraph (CCITT) to adopt the asynchronous transfer mode (ATM). The transfer mode is a collection of mechanisms that are used to implement switching and multiplexing in the network. The success of ATM networks depends on the development of effective congestion control schemes. These schemes are responsible for maintaining an acceptable quality of service (QoS) level that is deliverable by the network. The congestion control schemes will decide to accept or reject new connections based on their traffic characteristics and available network resources. The congestion control schemes will police the existing connections to insure that they do not exceed their negotiated traffic characteristic parameters. Performance modeling techniques are needed to determine which congestion control techniques should be used. Performance modeling techniques include: analytical techniques, computer simulation, and experimentation [1]. Performance models require accurate traffic models which can capture the statistical characteristics of actual traffic. If the traffic models do not accurately represent actual traffic, one may overestimate or underestimate network performance. This article surveys traffic models in telecommunication networks. Traffic models can be stationary or nonstationary. Stationary traffic models can be classified in general into two classes: short-range and long-range dependent. Short-range dependent models include Markov processes and Regression models. These traffic models have a correlation structure that is significant for relatively small lags. Longrange dependent traffic models such as Fractional Autoregressive Integrated Moving Average (F-ARIMA) and Fractional Brownian motion have significant correlations even for large lags. Traffic models are analyzed based on goodness-of-fit, number of parameters needed to describe the model, parameter estimation, and analytical tractability. To evaluate goodness-of-fit, one needs to define metrics that determine how “close” the model is to the actual data [2]. These metrics have to be directly related to the performance measures that are needed to be predicted from the model. The goodness-of-fit used in this article is based on the ability of the model to capture marginal distributions, auto-correlation structure, and ultimately predict delays and cell loss probabilities. This article is organized as follows: The second and third sections cover traditional traffic models, Markov, and Regression models. The fourth section discusses nontraditional traffic models. It briefly reviews long-range dependence and discusses three different long-range dependent traffic models, fractional Brownian motion, F-ARIMA, and aggregation of high-variability ONOFF sources. Finally, the fifth section concludes the article.