An estimation technique for cell loss ratio in ATM networks with bursty multiclass sources

Estimation of cell loss ratios in a multiclass ATM network turns out to be a computationally intensive problem. In this paper, a computationally efficient algorithm is developed through a series of approximations. Multiple classes of bursty input sources are first approximated into an equivalent single class model. Based on the single-class model, the relationship between the arrival process and the buffer sizes is established by way of a renewal process alternating between overload and underload periods. Using variables associated with the renewal process, a typical two-dimensional model involving the arrival process and the buffer sizes is simplified into a one-dimensional model. A further approximation is made by estimating the available buffer space at beginnings of overload periods. Numerical and simulation results show a good match at moderate buffer sizes.