Diffusion models to study nonstationary traffic and cell loss in ATM networks

We study the effects of nonstationary traffic patterns in a network of ATM nodes. Dynamic behaviour of ATM networks is of interest due to the highly nonhomogenous nature of the load: periods of basic activities are interleaved with bursty periods of demands. The models frequently used to predict transient behaviour of these networks are based on fluid approximation. Usually they assume Poisson arrivals and consider only mean values of queues. Here, we propose a diffusion model which takes into account general input process and allows us to study the dynamics of nonstationary traffic along virtual path, to approximate transient distributions of queues and transient distributions of response times of one or several nodes. It also permits the estimation of time-varying loss rates due to limited capacity of buffers.