Diffusion models of leaky bucket and partial buffer sharing policy: transient analysis

Leaky bucket and partial buffer sharing policy have already become classic examples of preventive and reactive traffic control functions implemented in ATM networks. They help to respect the negotiated connection parameters, to avoid the congestion and therefore to ensure the guaranteed quality of service. We revisit their performance models with the use of diffusion approximation adopting our previously developed method of transient state analysis and extending it to the case of state-dependent input. This kind of approach gives us an inside look upon the transient behaviour of the traffic. The dynamics of the traffic is displayed and the influence of both mechanisms on the traffic characteristics appears as a function of time. General cell interarrival times and the burstiness of the traffic are represented in a natural way in these models. The diffusion method is a second-order approximation and thus has certain superiority upon the fluid approximation. Both models can be easily implemented in a general queueing network model. Hence, the impact of both mechanisms on the performance of the whole network may be studied. The models may be applied also in cases of very small losses which are difficult to study by simulation.