> glob2 uses deterministic behavior that is expected to be simulated on all > clients identically, reducing the amount of network traffic to the human > actions. This approach certainly has its academic value but turned out to be > fatal as it bound the game speed to the slowest client's speed and caused > big syncing trouble when one client had connection issues. Also by its > nature all clients had to know all game state and only because people plaid > fair, the fog of war was not lifted above the enemy base, so yeah, consider > the current network design as obsolete. > > The only part that might not be obsolete is that games have to run off the > meta/handshake server once the game starts, meaning that one of the clients > should take the role of the server to be scalable. Alternatively servers > should be dedicated machines like it is common in FPS servers, but the meta > server should be able to handle many such servers dynamically.
In addition, the current network model for glob2 is a headache with respect to NAT traversal. We spent countless of hours in trying to make a network protocol able to traverse NAT. Although current NAT control protocol such as UPnP could help, I agree with Leo that glob2's network model is too much of a hurdle. The question of an alternative network model is for me open. have a nice day, Stéphane -- http://stephane.magnenat.net _______________________________________________ glob2-devel mailing list [email protected] https://lists.nongnu.org/mailman/listinfo/glob2-devel
