Influence of input flow character on network dimensioning

The character of the input flow changes with an increasing amount of multimedia services. The most common classical stochastic models suppose an exponential distribution of interarrival times, i.e. the Poisson input flow. This paper shows the significant influence of probability distribution functions of interarrival times, such as the exponential, the hyper-exponential and the Erlang's of the order k, on the loss probability of the multi-service system. These types of system models are often used for the capacity planning and network dimensioning. The presented analytical and simulation model supposes very specific character of input flow, which is formed from packet streams of interactive services with different bandwidth requirements. Simulation results for various interarrival time distributions are compared with the analytical solution of multidimensional Erlang's model. In case of exponential character of input stream the simulation results are in good agreement with analytical model. For other distributions of interarrival times significant influences of variation coefficient and service type are shown.