The call usage (expressed in Erlangs) between two facilities represents the average number of simultaneous bidirectional media streams between those facilities. For example, if the call usage between two facilities is 100 Erlangs, then the average number of simultaneous calls is 100. However, since this is only an average, roughly 50% of the time there are more than 100 simultaneous active calls. So, it would be a mistake to simply multiply 100 media streams by the appropriate value for kbps per stream.

The goal is to supply enough bandwidth to adequately support the media streams at least 99.9% of the time. Given a call usage, the Erlang B model is used to estimate the 99.9th percentile value for the number of simultaneous streams. For example, if the call usage rate is 100 Erlangs, the Erlang B model implies that there are at least 128 simultaneous media streams less than 0.1% of the time. So, in that example, it is sufficient to engineer the bandwidth to support 128 simultaneous media streams.

Once you determine the 99.9th percentile for the number of simultaneous media streams, it can be converted to kilobits per second by using the numbers in Typical LAN bandwidth requirements for media streams. For example, for a typical LAN configuration, the G.711 codec, and a packet size of 20 ms, Typical LAN bandwidth requirements for media streams implies that 87.2 kbps are required per call. In that case, the required bandwidth would be 87.2 kbps x 128 = 11.2 Mbps in each direction.