Efficient estimation of cell blocking probability for ATM systems

The assessment of the cell loss performance of networks using asynchronous transfer mode (ATM) via Monte Carlo simulation incurs an enormous computational burden due to the need to estimate an event that has a very small probability of occurrence. Although importance sampling (IS) techniques have been proven useful in simulations of rate events related to bit error rate in digital communications and false alarm rate in radar systems, its application to ATM queuing problems with correlated input traffic has yet to be demonstrated. It is established that significant computational savings can be obtained using IS for correlated traffic by using regenerative properties of the underlying system and biasing the conditional arrival process. The results show that IS can reduce the computational burden by more than three orders of magnitude. Extensions of the methodology to more complex arrival processes are discussed. The foundation for applying IS to ATM systems given can be used to study congestion control as well as networks of ATM queues in the future