Title: When to stop simulation
Simulation plays an important role in various research fields and provides a feasible way of solving analytically intractable problems. The research object is modelled via a computer programme. Different groups of parameters will be included in the code and the output of interest collected. Before a simulation starts, how long the simulator needs to be run is a question. The output data needs to be enough so as to ensure the accuracy of the simulation results. However, the simulation needs to be stopped as early as possible, so that it will not run an unnecessarily long time, wasting time and computer resources. Simulation run length prediction is therefore an important question, which should not be neglected when planning any simulation. Most of the researchers in this area run a pioneer simulation to try to get a rough idea when to stop the simulation. However, different groups of parameters settings might get different results. Therefore, the accuracy of the output cannot be guaranteed. This PhD research focuses on predicting the simulation run length of packet-queue based network models using an analytical approach. Our target is to find a formula to calculate the required run length. As is well known that run length prediction is model-dependent, this research studied a well-accepted queue model: a multiplex of multiple Markovian ON/OFF sources into a queue.Results show that the required run length can be predicted using this approach, and we can provide a guideline for how to solve such a problem for other queue models.