Simulation: Scaling Processing Power

When thinking of simulation within a game it is usually within the confines of a single computer system. However, what if your processing needs might exceed the user's machine, but you still want to create the detail of a much bigger set of processing power?

I was flipping through the book "Tricks of the Game Programming Gurus". This is an older book and most of the information is outdated, but one thing in there sparked a thought. It was the subject of AI. It talked about how in the really old computer game days programmers would come up with "patterns" for the computer AI. There was jut not enough processing power to display the game and do deep AI "thinking". It was later that it hit me: Why can't we offload the processing power for much deeper AI decisions to a machine on the internet? This is what an MMPORG is doing at some level.

Now for practicality and speed some things will not easily go off the machine. Such as say a vehicle simulation there would be a speed issue. For AI, however, a significant amount of depth could be achieved. The speed would not be critical other than the bandwidth for sending enough information to make decent decisions.

Where am I going with this? The current project I am cooking up I am wanting to do a simulation with no human interaction except for decisions as to characters and equipment. The simulation will be done weekly with results posted for users to check them out. One thing I wanted to do was allow a lot of AI in the simulation. The engine will most likely be loaded just trying to display the large number of bots in the game. The video part is of importance to me because I intend to make this a spectator sport. So if I want really sophisticated AI group thinking and individual thinking I started thinking of options. That is when I found this: www.parallelpython.com/ It was the result of a conversation I had with someone here at GG about threads. It really intrigued me.

I have come to understand that a lot of super computing power is controlled and synced with a Python interface. Not that the language is important, but it would allow a quicker development time to leverage those resources. Then I thought if you have a dedicated process running in a thread next to the desktop or server Torque simulation then you could use it to coordinate external computing sources: programs on other cores or on other machines. That means you could hook in an Amazon.com super computer cluster for specific simulations only paying for the time you use the cluster. Now this may not be really practical for an MMPORG to constantly use such a resource so purchasing a cluster and maintaining it might be better in that case. However, who knows, in the future it may be less costly to purchase that kind of computing power even if using it constantly. It certainly would be more conducive to scaling.

Now I am just thinking about my own project and renting some high end computer resources each week to do highly resource intensive simulations sounds like a practical thing for my goals. I hope this inspiration has the affect of getting people thinking about a potentially useful resource for game simulation. It certainly triggered some cool things in my mind as to what I could do with it.