Recovering the traffic wave from counts alone

Network traffic does not flow at a constant rate. There are busy periods and quiet ones, and the variation is correlated over time. The stochastic model that captures this wave is the Markovian arrival process (MAP).

The difficulty is that measurements usually come already aggregated — “so many events per minute”. Without the individual arrival times, conventional estimation methods cannot be used.

This work gives an EM algorithm that estimates the parameters of a MAP from grouped data alone. It is formulated so that the computation does not break down when records are missing, which makes it usable in the practice of network performance evaluation.