Statistical models for long term network traffic in enterprise networks

Abstract

With the rapid development of the internet it has converted the world into a global village and now a day we cannot even think of a micro second down time. For an instance, user demand has caused the internet to successfully combined with other networks. This expanded development has caused for huge internet traffic loads and network congestion. For solving this key issue of the networks it is important to predict the traffic peaks in the network. These traffic peak is caused by a large amount of data being requested like in a download. If these traffic peaks are predictable then non critical traffic from another network can be scheduled to avoid peak to reduce the congestion and maximize utilization. This dissertation introduces a method to solve that key issue. Curve fitting technique in Matlab and distributing fittings are used to build statistical models of predicting traffic. Once that identifying some drawbacks through curve fitting methodology it has been rejected and statistical models for long term network traffic in Enterprise network is used as the proposed technique. Pareto distribution, Beta-Prime distribution and Exponential distribution are derived as the statistical models to predict the traffic peak in Enterprise network. The analysis is conducted by looking at the predictability of a peak in terms of level crossing of a given level. According to the available literature there was no such technique for predicting traffic peaks. As per the results curve fitting methodology error is significantly high. Beta-Prime and Exponential distribution are not good statistical models of predicting traffics due to huge error occurred when compared to the actual behavior of the network. But Pareto distribution is the best model of prediction on traffics in the network as it has vey less error when compared to the actual behavior of the network. According to the results Pareto distribution is the best statistical model of predicting traffic peak. Once predicting the traffic peak can be scheduled the large data from other network for maximum utilization and to avoid the traffic congestion.