Title : Duration of simulations
Author: J. Charzinski
Determining the appropriate duration of a simulation is a critical issue, especially when rare events may cause drastic effects. Examples are (1) the well-known problems of heavy-tailed distributions in general, but also (2) multi-level simulations where care must be taken to simulate a sufficient number of events also on the most coarse time scale and (3) simultions of protocol state machines like exponential back-off where some protocol states are rarely reached but cause extreme delays. Depending on the way measurements are collected in a simulation, these cases may even be invisible in the results because e.g. when latencies within a connection that does not end before the end of the simulation run are not counted in the statistics.
Download as PDF