Characterisation of busy-hour traffic of IP networks based on their intrinsic features

Characterisation of busy-hour traffic of IP networks based on their intrinsic features is a scientific paper that discusses the use of statistical techniques to model the behaviour of busy-hour traffic, especially to give network designers and architects grounded data on how to plan for a network based on the critical measure of peak network use. The main experiment conducted for this paper consisted of observational traffic analysis, with the data rate measured during busy hours. The solution explored an architecture that consisted of an initial stage for data cleaning followed by statistical regression. This latter step was modelled to find the relationship between two variables: number of users (predictor variable) vs traffic volume (target variable). These two variables are analysed to check whether there is a real relationship between them regardless of physical location or link capacity. It eventually came to light that the analysed traffic, which stemmed from a number of educational institutions and local networks, follows a White Gaussian Process, meaning that data us uncorrelated, comprising of random and independent fluctuations. What also entails from this observation is that the samples of traffic distribution follow a normal (bell-shaped) curve. Whereas a normal process would reveal that traffic in the real world would be coloured, with clearly defined correlations, a White Gaussian process indicates the presence of noise, meaning that the variables are unrelated due to presence of noise. A White Gaussian Process can also make for a simplified thought process which relies on simple math to predict probabilities of the network being overwhelmed, like predicting the network throughput in conrete numbers e.g.: gbps where the traffic rate becomes throtelled. Lastly, it has been concluded that the relationship between user population and traffic use can be modelled by the ANOVA and ANCOVA analytical techniques. The former stands for analysis of variance, which consists of a test to check whether the means of distinct groups are significantly different. The latter technique handles covariance, which couples ANOVA with regression, detailing how a dependent variable changes according to a determining factor while controlling for another factor. The use of tThese two techniques were arrived at through Goodness-of-fit tests, namely, tools to confirm that the traffic variation was mlinked to the number of users. Ultimately, the authors were able to fit the model to a linear regression analysis, derivating the formula for traffic: (traffic per user) x (number of users) + error term.