How is a real-world simulation designed?

Question

I am fascinated by the performance of applications such as "Rollercoaster Tycoon" and "The Sims" and FPS games. I would like to know more about the basic application architecture. (Not so concerned with the UI - I assume MVC/MVP piriciples apply here. Nor am I concerned with the math and physics at this point.)

My main question deals with the tens or hundreds of individual objects in the simulation (people, vehicles, items, etc.) that all move, make decisions, and raise & respond to events - seeming all a the same time, and how they are designed for such good performance.

Q: Primarily, are these objects being processed in a giant loop, one at a time - or is each object processing in it's own thread? How many threads are practical in a simulation like this? (Ballpark figure of course, 10, 100, 1000)

I'm not looking to write a game, I just want the design theory because I'm wondering if such design can apply to other applications where several decisions are being made seemingly at the same time.

Answer 1

There are two basic ways of doing this kind of simulation Agent Based and System Dynamics. In and agent based simulation each entity in the game would be represented by an instance of a class with properties and behaviors, all the interactions between the entities would have to be explicitly defined and when you want these entities to interact a function gets called the properties of the interacting entities gets changed.

System Dynamics is completely different, it only deals with sums and totals, there is no representation of a single entity in the system. The easiest example of that is the Predator and Prey model.

Both of these have advantages and disadvantages, the System Dynamics approach scales better to large number of entitities while keeping runtime short. While there are multiple formulas that you have to calculate, the time to calculate is independent of the values in the formula. But there is no way to look at an individual entity in this approach. The Agent based approach lets you put entities in specific locations and lets you interact with specific entities in your simulation.

FSMs and Celular automata are other ways in how to simulate systems in a game. E.g. in the agent based approach you might model the behavior of one agent with a FSM. Simcity used Celular automata to do some of the simulation work.

In general you will probably not have one big huge model that does everything but multiple systems that do specific tasks, some of these will not need to be updated very often e.g. something that determines the weather, others might need constant updates. Even if you put them in separate threads you will want to pause or start them when you need them. You might want to split work over multiple frames, e.g. calculate only updates on a certain number of agents.