A Versatile Probability Distribution for Light and Heavy Tails of Web File Sizes

A novel unique probability distribution, which has a lognormal body and either light or heavy tail, has been fitted to various empirical data sets of Web file sizes. The optimal parameters of this distribution have been determined by the maximum likelihood estimation combined with the optimization algorithm minimizing a goodness-of-fit metric specially adopted to provide the best fit to the upper tail. The mirror transformation of the processed original data set with respect to the median has been proposed to improve the fit. The obtained results question the common opinion that the probability distribution of Web file sizes is heavy-tailed. The lognormal fits are given for comparison.