The paper recommends an approach to estimate effectively the probability of buffer overflow in self-similar queuing networks with finite buffer capacity. Simulations with stochastic traffic source models are conducted. A new efficient algorithm, based on the RESTART method with the Limited Relative Error (LRE) technique, is developed and applied to accelerate the buffer overflow simulation in a finite buffer single server model under long-range dependent self-similar traffic load with different buffer sizes. Numerical examples and simulation results are shown.