Hi! I've read one or two papers about the client side prediction technique, but I have a fundamental question that isn't answered in any of them: If on the server side objects are simulated in lockstep with the client, that is, in bursts of steps always when the next client input arrives, what is the best way to handle interaction between client avatars on the server? There is no common concept of time, so there must be some kind of compromise. Before starting out and trying all kind of things I would appreciate some input from people who have already gathered some experience with this problem. Regards cm

The server should always have the authoritative state, so it will know where those two characters 'actually are' and can do the interaction accordingly, sending messages to both (all?) clients informing them of what has happened.

For car crashes: When the local car sees a collision, it can add a bunch of smoke, which means that you won't see anything while the server is figuring out what to do :-)