Advertisement

Getting players to make rooms without port forwarding

Started by April 13, 2014 01:36 AM
2 comments, last by hplus0603 10 years, 7 months ago

Hello.

I've always wondered this on my mind and I've had a hard time looking for papers or articles on how this is done exactly.

Take for example the game: GunZ:

http://mmohuts.com/wp-content/gallery/gunz/gunz-game-lobby.jpg?ec9f9b

In this game you can connect to a certain number of servers after you've logged in.

Once you're logged in, you're connected to the lobby where you can chat with other players

connected to the main server.

However, you can also create your own "rooms" and "host" your own game for others to play on.

Now it's not 100% for sure that anyone here may know how exactly GunZ does it, but does anyone have an idea

on how this actually works?

When a player hosts a room, and someone wishes to join your room, are other people connecting to that player?

But that player hasn't even port forwarded and they're managing to connect to him anyways, are they using some sort of

UDP holepunch / NAT traversal?

Are players technically actually hosting their own servers for other players to connect to? Because if that's the case isn't this a huge security risk since they can simply just edit memory on their server?

If they aren't indeed hosting their own servers (rooms), then does that mean every single room on the server is being emulated through the main server? Meaning when you connect to a players room, you're still actually sending datagrams/packets to the main server?

I don't know how GunZ does it. I can think of two ways of doing it:
- If serving the game is cheap in CPU and RAM resources, then a new "room" would just be another instance on the server -- it could be in-process, or a new process.
- If serving the game is not cheap enough to do that, then it's likely NAT traversal/hole-punching is used, and the server process runs on the user's machine.
enum Bool { True, False, FileNotFound };
Advertisement

I don't know how GunZ does it. I can think of two ways of doing it:
- If serving the game is cheap in CPU and RAM resources, then a new "room" would just be another instance on the server -- it could be in-process, or a new process.
- If serving the game is not cheap enough to do that, then it's likely NAT traversal/hole-punching is used, and the server process runs on the user's machine.

When you say instances in-process, are you talking about instances of some sort of "server" class?

And obviously 'new process' I'm assuming would be just multiple processes of servers.

But would you be using the same socket for each? Or different sockets?

What about ports? Would each "instance" be using a different port?

What if the game was insanely popular and it had a TON of rooms (hundreds) of rooms.

Wouldn't the server need to port forward all of those ports if they all use different ports? And if so, isn't that a huge security risk?

In your other case, where the players themselves have the server running through a NAT holepunch method, what exactly can a developer do to prevent the players from manipulating data in their servers to better themselves in the game? (e.g. increasing their score)

But would you be using the same socket for each? Or different sockets?
What about ports? Would each "instance" be using a different port?


With UDP, you want exactly one socket per port. If you fork to create another process, you want another port for that, meaning that clients will have to be told to connect to another port. Alternatively, you could have a single "gateway" process that just receives the data, and forwards it to the appropriate process for processing, and then responds back with whatever that process returns. (Erlang has some nice built-in functions for this kind of thing, btw.)

With TCP, you can inherit connections (sockets) across fork in a useful way. If you still want to be connected to both the master and the room, then you still need another connection for the new server process, which in turn means you probably want to allocate a new port for establishing that connection. Or, again, use a gateway of some sort.

If the new rooms are "in process" (yes, some kind of server object, or other state kept in the same process) then none of that is necessary; the "gateway" part is just the part of the code that reads incoming data and passes it to the appropriate code part for processing.

For a game with 100s of rooms, again, it matters whether serving is simple or hard. On a modern server, competently multi-threaded, with event/asynchronous I/O for networking, I would expect to be able to run hundreds of eight-player Quake-I-style matches (sliding spheres against BSP trees for the "physics" part) without degradation. If you need more than that, yes, you're going to have to start splitting onto multiple server hosts (machines.)
enum Bool { True, False, FileNotFound };

This topic is closed to new replies.

Advertisement