May 13, 2009 - Persistence of vision
Last night I began to put a polish on the classes I’ve been labouring over for the past few months. The reason I’ve been labouring is that I’m trying to write this game in a way that makes netplay easy to implement. One of the biggest hang-ups has been on the role of a server in the whole thing, and how the data gets to and from it.
A big problem with HTTP and subsequently most of the technology designed around it is the lack of persistence. Simply put, when you visit a webpage, the server will fire up its processing engine, initialise all its variables and objects, read stuff from the database, process your request, then clean up after itself. This is fine for 99% of the things you’d do on the web, but when making a semi-realtime game run over this system, it becomes horrendously inefficient. By contrast, most games will load all the data into memory and it remains there in a persistent state, occasionally being altered by the course of the game.
So, if you want to make a game that runs online, you have an extra layer of complexity to think about in that you have to ‘serialize’ the game’s state into a format that can be sent over the internet to the server, then the server will ‘deserialize’ the data, process it, save it to its database, and serialize a response to send back. There are a few APIs available which handles these sort of operations quite well, but it’s still a complex thing to monitor.
Of course, you could do away with the server altogether (although this isn’t usually an option for a web app), or reduce its task to simply routing the data to wherever it needs to go. There are a few big problems to tackle with this however:
- Cheating – without a server, this means every client is doing its own verification of valid moves etc. Someone could fairly easily exploit this and send invalid moves.
- Synchronisation – with no centralised game state to reference, clients could get out of synchronisation.
The answer to both of these problems would be to perform validation on all the clients, so that everyone’s client would have to agree the move is valid within the context of their stored game states in order for it to be passed. This has its own problems of course, as it’s dependent on latency and consistency between client versions, and may make the game feel slow and laggy.
So after much deliberation about how to do it, I think I’ve finally found a way to handle server-driven games. The abstraction between the game data format and the renderer means that I have a very clean set of objects to represent games. By recreating this in C++ or Java on the server side, and using JSON to represent the objects and their properties as they’re transmitted from client to server and back again, I can build up a persistent remote version of the game on the server. Once a game is running, I don’t have to worry about storing and retrieving the data across requests, which eliminates much of the overhead. So the sequence of events that occurs when a Horse moves from tile A to tile B is as follows:
- The server tells all players that the game is in a movement phase, and tells player 1 it’s his turn
- Player 1 selects his Horse and tells it to move to tile B
- Player 1’s client works out the path to tile B (it’s more than one tile away so it requires the most efficient to be worked out)
- Happy with the path, the client sends the server some data, namely the player ID, unit ID and each step of the path in sequence. It also presumes the move to be correct and shows the updated position (this is called client prediction, and is very important in online gaming to produce a responsive interface – the chance a move will be legal will be many, many times more than the chance a move will not)
- The server receives the request and checks against its own version of the game to see if the unit can make this move legally; it double checks the path.
- Having verified the move is legal, the server sends back the player ID, unit ID and its final position.
- Player 1’s client receives the validated move.
Of course, if the move is invalid, the server will send back a rejection, with the player ID, unit ID and its original position as defined on the server. The client will then move the unit back and display an error. It should be noted that invalid moves will be next to unheard of as long as the server and clients are functioning correctly, and are there essentially to combat cheating.
Finally, to combat disconnections or to simply allow the game to be played over a long period of time and over multiple platforms, the server can send an entire representation of the game to the client to get it up to speed.
In closing, it’s a lot of effort to get a game to work over the internet, but it’s worth it to be able to live out that long held fantasy of playing Chaos with your mates – no matter where they are…