Question

I have some experience making multiplayer turn-based games using sockets, but I’ve never attempted a realtime action game. What kind of extra issues would I have to deal with? Do I need to keep a history of player actions in case lagged players do something in the past? Do I really need to use UDP packets or will TCP suffice? What else?

I haven’t really decided what to make, but for the purpose of this question you can consider a 10-player 2D game with X Y movement.

Answer 1

• ‘client server’ or ‘peer to peer’ or something in between: which computer has authority over which game actions.

With turn based games, normally it’s very easy to just say ‘the server has ultimate authority and we’re done’. With real time games, often that design is a great place to start, but as soon as you add latency the client movement/actions feels unresponsive. So you add some sort of ‘latency hiding’ allowing the clients input to affect their character or units immediately to solve that problem, and now you have to deal with reconciling issues when the client and servers gamestate starts to diverge. 9 times outta 10 that just fine, you pop or lerp the objects the client has affected over to the authoritative position, but that 1 out of 10 times is when the object is the player avatar or something, that solution is unacceptable, so you start give the client authority over some actions. Now you have to reconcile the multiple gamestates on the server, and open yourself up to a potentially ‘cheating’ via a malicious client, if you care about that sort of thing. This is basically where every teleport/dupe/whatever bug/cheat comes up.

Of course you could start with a model where ‘every client has authority over ‘their’ objects’ and ignore the cheating problem (fine in quite a few cases). But now you’re vulnerable to a massive affect on the game simulation if that client drops out, or even ‘just falls a little behind in keeping up with the simulation’ – effectively every players game will end up being/feeling the effects of a lagging or otherwise underperforming client, in the form of either waiting for lagging client to catch up, or having the gamestate they control out of sync.

• ‘synchronized’ or ‘asynchronus’

A common strategy to ensure all players are operating on the same gamestate is to simply agree on the list of player inputs (via one of the models described above) and then have the gameplay simulation play out synchronously on all machines. This means the simulation logic has to match exactly, or the games will go out of sync. This is actually both easier and harder than it sounds. It’s easier because a game is just code, and code pretty much executes exactly the same when it’s give the same input (even random number generators). It’s harder because there are two cases where that’s not the case: (1) when you accidently use random outside of your game simulation and (2) when you use floats. The former is rectified by having strict rules/assertions over what RNGs are use by what game systems. The latter is solved by not using floats. (floats actually have 2 problems, one they work very differently based on optimization configuration of your project, but even if that was worked out, they work inconsistently across different processor architectures atm, lol). Starcraft/Warcraft and any game that offers a ‘replay’ most likely use this model. In fact, having a replay system is a great way to test that your RNGs are staying in sync.

With an asynchronus solution the game state authorities simply broadcast that entire state to all the other clients at some frequency. The clients take that data and slam that into their gamestate (and normaly do some simplistic extrapolation until they get the next update). Here’s where ‘udp’ becomes a viable option, because you are spamming the entire gamestate every ~1sec or so, dropping some fraction of those updates is irrelevant. For games that have relatively little game state (quake, world of warcraft) this is often the simplest solution.