Statistical Modeling of Internet Traffic FlowLength and Flow Size

The growth of internet traffic forces telecom service providers to invest in new infrastructure and/or expanding the network to achieve the desired Quality of Service (QoS) and cope up the network congestion. But several techniques available for granting QoS and network congestion. The first step is to understand the network performance. Network performance needs accurate traffic modeling that has the potential to improve desired QoS, allocating the network resources (i.e. bandwidth). This thesis presents a statistical model of the internet traffic applications by their random nature of flow-length and flow-size. Wireshark network monitoring and analysis tool is used to collect internet traffic data from the ethio telecom core switch and generating experimental internet traffic data in controlled environment. The experimental data are used to train and test the machine learning model that helps to identify the internet applications. Firstly, identifying internet traffic applications using machine learning classification techniques. Secondly, statistical methods are used to fit the Cumulative Distribution Function (CDF) and select the parameters that best fitted in both flow-length and flow-size for identified applications. Finally, deliver statistical model to each applications and corresponding parameters. Recently internet traffic modeling is applicable in capacity planning for traffic engineering, anomaly detection and performance analysis are some of them. Based on the result found the Log-normal distribution is best fitted to flow-length and flow size for three applications and Weibull distribution is for SSH application in both flow length and flow size.