Traffic engineering experience in a nation-wide cellular service network

We re-examine current traffic engineering criteria in a nation-wide cellular service network supporting more than 15 million subscribers. Call blocking rate is measured at the links connecting the MSC (mobile switching center) and CGS (cellular gateway system). Contrary to the common belief that the shortage of link capacity is most responsible for call failure (failure to setup an end-to-end circuit between a calling cellular phone and the callee), it accounts for only 36% of the total occurrences of call failure. This implies that merely increasing the link capacity might not necessarily mean zero call failure. We also found that the measured blocking rate tends to be higher than the expected blocking rate from the Erlang B table. Short-range traffic characteristics, obtained from billing data at the fine resolution of seconds, clearly resembles a Poisson distribution, the basis for the Erlang B table. However, long-range traffic characteristics, obtained by collecting busy hour traffic over a 3 month period, shows large deviation from a Poisson distribution. Variance mean ratio (VMR) is as high as 6.71 (unusually high and low traffic on special events or weekends are excluded). This is rather higher than expected, since a true Poisson traffic would exhibit VMR of 1. Based on this observation, we recommend a greater safety margin when determining the number of trunks for a cellular service network.