How to run efficient simulations to analyze real-world systems

To perform a correct simulation, the simulation software must be properly structured. The program structure must consist of three software levels:

• The simulation executive software controls the execution of the model program. It sequences operations and manages modularity problems by separating generic control from model details. Two techniques are possible: the synchronous execution technique and the event-scanning technique.
• The model program implements the model of the system to be simulated and is executed by the simulator.
• The routine tools generate random numbers by deriving them from typical probability distributions to obtain statistics and to manage problems related to modularity by separating generic functions from those specific to the model.

The main cycle of a simulation generally consists of three phases:

• Start phase: The termination condition is initialized to false; the status variables are initialized, the clock is set to the simulation start time (usually zero), and the initial events are placed in the event list.
• Loop phase: The cycle continues until the termination condition occurs. At each step, the clock is set equal to the start time of the first event in the queue, the event is simulated and removed from the queue, and finally, the statistics are updated.
• End phase: The simulation ends by generating a report with the statistics of the simulated system.

The simulation can be applied to all systems that can be modeled, respecting the characteristics mentioned in the previous paragraphs. This approach is particularly suitable for diagnosing problems related to complex processes. Understand where the system bottlenecks are. These areas are critical as they significantly decrease system performance and are usually associated with slower components. The only way to significantly improve the performance of a process is to improve it precisely in these critical areas, which, unfortunately, in many processes, are clouded by excess inventory, overproduction, diversity of processes, different sequencing, and stratification.

Therefore, by modeling these processes accurately and inserting them into a simulator, it is possible to obtain a more detailed view of the entire system, find the bottlenecks, and use performance indicators to analyze the system and improve its performance.

A simulation is a numerical model that mimics the operations of a real existing system, such as the daily operations of a bank, the process of a factory assembly line, and the assignment of personnel in a hospital or call center. For this reason, it must consider all the resources and constraints related to the system and their interactions over time. It must also coincide as much as possible with reality and therefore consider that events can have variable completion times, and this is done by introducing pseudorandom generators. In this way, the simulation is much closer to the real situation, and therefore through it, it is possible to predict the behavior of the system in the face of changes in the input data or its structure, just as if the simulation were real. Therefore, with this type of simulation, you can test your ideas much faster and at a much lower cost than real simulation.

Generally, any real system that is expressible through a flow of processes with events can be simulated. The processes that can be most beneficial are those with more changes over time and with a high level of randomness. A good example is a vehicle filling station: you cannot predict exactly when the next customer will arrive, nor the type and quantity of fuel they will require.

Consequently, the simulation, where applicable, should be preferred to the real simulation since a gain in terms of cost, time, and repeatability of the simulation is obtained. Just think of the cost associated with each real simulation, not only as regards the expense of hiring new employees or purchasing new components, but also the consequences that these decisions can have on the real system, which can lead to positive but also negative results. On the other hand, this is avoidable in numerical simulation in which the system has no connections with the real system, and therefore the test does not cause economic consequences.

Furthermore, it is more difficult to simulate the same real system twice with the same conditions, while in numerical simulation, you can test a system with the same conditions by modifying only the input. This gives greater certainty that one idea is better than another. Finally, as mentioned before, the time needed to carry out a numerical simulation is much shorter than that of a real simulation, especially if the simulated system takes a very long time. For example, if we want to analyze the throughput of customers in a month, the real simulation must pass at least 1 month, and the numerical one just a few seconds.