Effect of on/off distributions on the cell loss probability in ATM networks

The effect of the distributions of on/off periods (of on/off sources) on the cell loss probability in an ATM network is discussed. Based on the peak and average rates and/or the average burst lengths of the sources, two-state on/off sources, whose on and off periods are exponentially or geometrically distributed, are then constructed. Decomposition methods and asymptotic analysis are used to derive simple, accurate approximations for the cell loss probability in a statistical multiplexer with a finite buffer. Numerical results indicate that, for a given fixed burst length, the distributions of the on/off periods have a significant impact on the cell loss probability. The approximate results are verified using simulation results and are found to be very accurate for most cases. For a buffer size of 400 cells, the maximum link utilization that the link can achieve subject to a cell loss probability of 10/sup -9/ is found to be only about 50-60%. Therefore, heavy load approximation should be used cautiously. The results presented can be used in admission control.<>