Statistical properties of PMA traffic traces and their application to capacity sizing
Zdenko Vrdoljak, Marije Ljolje · 2005
In this paper we describe statistical properties of a real network traffic data set and the capacity calculation formula based on these properties. These are results from the joint project of the FER, University of Zagreb and Ericsson Nikola Tesla, Zagreb, with the aim to analyze statistical properties of publicly available IP traffic traces and study the influence of these properties to traffic sizing procedure. The analysis is based on the network traces database provided by the Passive Measurement and Analysis (PMA) project of the National Laboratory for Applied Network Research (NLANR) of USA. We describe traffic rate distribution fitting procedure. The distribution can be closely described by the sub-exponential distributions and that the time-dependence is strong. The most important result is that the packet delays observed on different nodes in the network belonging to the same traffic flow can be analyzed as (highly) independent random variables. This is especially true when the capacity of the links in the network are chosen in such a manner that a maximum of 1% of all packets have a delay higher than 10 ms. We describe a fully empiric capacity calculation formula that can be applied to all links in the network knowing the traffic matrix. The formula is based on the statistical properties established before. The formula takes into account only the mean value of the traffic rate which makes it very simple to use.