Simulation modeling, experimenting, analysis, and implementation

Textbooks sometimes describe building models, running experiments, analyzing outputs, and implementing results as distinct activities in a simulation project. This paper demonstrates advantages of combining these activities in the context of system performance optimization. Simulation optimization algorithms can be improved by exploiting the ability to observe and change literally anything at any time while a simulation is running. It is also not necessary to stop simulating candidates for the optimal system before starting to simulate others. The ability to observe and change many concurrently running simulated systems considerably expands the possibilities for designing simulation experiments. Examples are presented for a range of simulation optimization algorithms including randomized search, directional search, pattern search, and agent-based particle swarm optimization.