Packet size process modeling of measured self-similar network traffic with defragmentation method

Analysis and modeling of telecommunication networks by simulations has become one of the main tools in the process of telecommunication-networks' planning and upgrading. Knowledge regarding the statistical modeling of network traffic is very important. Here we tend towards modeled network traffic which would be the best possible approximation of the measured traffic. Throughout our research in the field of self-similar network traffic we have faced problem of statistically describing the packet-size process. We have noticed that small discrepancies between measured histograms and estimated probability density functions, as used in traffic generator models, lead to large discrepancy between measured and modeled network traffics. In this research we tried to estimate the probability density function of a measured histogram for process-packet size, in such way that would decrease these discrepancies. For this purpose, we have developed a novel method of modeling network traffic, which is based on the defragmentation of measured traffic. Using this defragmentation method, we can estimate parameters of files' size process, from captured packets and use these statistical parameters for traffic generation, via the OPNET simulation tool. From these simulations, we can show that this newly-developed method decreases discrepancy between packet size process histograms of measured and simulated network traffics. This consequently leads to a decrease in discrepancy between measured and simulated network traffics.