Throughput analysis of input-buffered ATM switch

The limiting throughput of an ATM (asynchronous transfer mode) switch with input queueing under the FCFS (first-come-first-served) scheme is 0.59. Three policies are proposed in the literature to improve the throughput of the switch: policy A (input expansion policy), policy B (window selection policy) and policy C (cell discarding policy). The author considers, for policy C, correlated traffic where all cells belonging to a burst are routed to the same output port and each traffic source is modelled by an interrupted Bernoulli traffic process (IBP). The results for policy C show that the destination correlation does not affect the throughput, and that the throughput is higher when all input traffic is balanced. Better results for policy B are obtained compared with the simulation results found in the literature. Results show that policy A achieves the highest throughput. The throughput obtained by policy A is higher by 0.07 than that obtained by policy B. For example, with expansion size s=8, the achievable throughput obtained by policy A is 0.94. With window size w=8, the throughput obtained by policy B is 0.87. The throughput obtained by policy C exceeds 0.8 when the input traffic rate is less than 0.5. The throughput, for policy C, is ρ=1-(1-p/N)^N whether the input is an IBP or Bernoulli and whether there is destination correlation or not. Simulation data, which are available in the literature for policy B, show better agreement with the author´s analytical results than with other analytical results in the literature