On the Reduction of the Complexity of Realistic Large Scale Network Simulations

Current computer systems allow a realistic simulation with more than 100,000 HTTP/TCP clients, as shown in this paper. However, the complexity of such simulations is high: the required memory and the simulation duration touches reasonable limits. Reducing the complexity by using a smaller number of clients as compared to reality is evaluated in this study. The reduction is performed with increasing the activity of each client keeping the load approx. constant. Further, it is shown that the average number of active clients remains approx. constant. The reduction has two targets: (i) the optimisation of the considered simulation scenario and (ii) to allow for simulations with larger simulation scenarios.

It is evaluated how the reduction affects the following parameters: average traffic load, coefficient of variation, Hurst parameter, end-to-end delay and loss probability. It is shown that only the loss probability is affected by the reduction. The simulation results show that the required memory can be reduced by a factor of 4–8, depending on the error bound, and the simulation speed increased by up to 33%. The gain allows to simulate an equivalent of 1,200,000 instead of 150,000 clients.