Research On Network Flow Model And Trace Forecast Technology

With the development of Internet, computer network changes people’s lives.Meanwhile, Internet itself becomes more and more complex. In the presence of thecomplex network, we need some ways to understand it deeply. Network measurementby which people grasp the structure and properties of the complex networks is animportant research direction. As two very important indicators we look upon veryimportant in network measurement: high bandwidth and low latency are verymeaningful. However, due to P2P traffic takes up much more Internet bandwidth, andcauses higher delay even network congestion, so we need some methods to controlnetwork traffic. In this case, if we can analyze the flow characteristics in the time scaleand get the prediction model, then we can control network traffic trends which areextremely important on network systems analysis and design in the future.This paper is the study of self-similar traffic model in network measurement. First,the paper describes some indicators of network performance and development processof network measurement model, and then explains that the current self-similar modelhas become the mainstream of current research models. Second, the definition andproperties of self-similarity are both illustrated. The paper proposes a variety ofmeasure methods of self-similar, and describes many self-similar models, then comparethe various models. According to the comparison, I choose FARIMA(p,d,q) modelwhich is adapted to the following experiment, and use it to fit the sequence of realnetwork traffic. The paper adopts the BIC method for estimating parameters of MA andAR, and results in the best match. Finally, Durbin-Levinson algorithm is used to predictthe flow for obtaining sequence of future flow fluctuations in95%confidence interval.The experimental procedure proposed effective measurement method ofself-similar network traffic sequence, and generates FARIMA(p,d,q) fitting modelwhich adapts to specific network successfully, and then arrive at a more reasonablefluctuations trend of network traffic in some lags. This method can effectively analyzedevelopment of a specific network without major interference, and has greatsignificance in the network flow control and congestion control.