MODELING OF BURSTY TRAFFIC USING HETEROGENEOUS ON-OFF SOURCE MODEL
واصفات البياناتعرض سجل المادة الكامل
Until recently it has not been clear whether Markov based models could be used to model bursty traffic. It has been claimed that the large number of states needed to model Journal of Engineering Research Issue (9)March2008 70 the traffic makes Markov models inapplicable for all practical purposes. This has initiated the search for other models that might be more suitable for modeling bursty traffic such as Fractional Gaussian Noise (FGN), Fractional Brownian Motion (FBM), Fractional Autoregressive Integrating Moving Average (F-ARIMA). For these models, however, the analytical tools for analyzing queuing behavior do not exist. However, they may be used in simulation. The ON-OFF source model is the most popular model for voice. It was used to model video traffic based on the minsources approach by Maglaris. Anick, Mitra and Sondhi used the ON-OFF sources to analyze bursty traffic. The ON-OFF source model is tractable for analysis when the transitions from the ON state to OFF state and from OFF state to ON state are exponentially distributed. In this paper, we will use classes of heterogeneous ON-OFF sources to model video data. This model is based on matching the total covariance of the heterogeneous sources to the real data. The covariance ofthe heterogeneous sources is composed of different exponential functions, while in the homogenous case it is just one exponential. The model is very attractive, because aswe will see for a small number of ON-OFF sources it is possible to get good results for the covariance and Index of Dispersion for Count (IDC). Moreover, the small number of parameters makes the analysis in finding the covariance and the parameters of the sources simple. The model we developed is different from that of Andersen, et. al. We used the Feldmann algorithm for approximating a long-tail covariance function by a finite mixture of exponentials. However, Feldman, et. al., used the algorithm to fit probability distribution. Our model is simpler than Anderson’s model. The matching of the covariance and for the real data to the traffic generated using the model is quite good.