An efficient computer simulation-based approach for optimization of complex polling systems with general arrival distributions

This study proposes an efficient computer simulation approach for estimation and optimization of performance measures in a polling system. A single server polling system operating under exhaustive, gated, and mixed service disciplines is developed. In this system, the arrival process is a Poisson process and service and setup times are exponentially distributed. The polling model is solved through two different methods: an exact method that requires the complete characterization of the system, and a computer simulation-based solution that reduces the solving time and the complexity of the model. A set of numerical experiments are presented in which it is shown that the computer simulation model outperforms the exact method in terms of estimating a system's performance measures. Moreover, it is shown that the optimizer simulation model is capable of handling general distributions and several queuing systems, whereas the exact method requires the complete characterization of the system through a Markov chain, which is a time-consuming and inefficient approach. In addition, the efficient computer simulation-based solution could be easily applied to polling systems with different numbers of queues and service disciplines.