There is a classic paper, which claims that many parameters of network traffic are best modeled by stochastic distributions with infinite variance. We believe that although this paper is correct, it has commonly been overinterpreted. It is true that some distributions have infinite variance, but not all. There are important cases where distributions of traffic parameters have finite variance. As an example, we show that the end-to-end available bandwidth must have finite variance, outline a method for measuring it, and present some experimental data.