I'm trying to follow the code given here to implement NAT hole punching in Node.js. I'd like to know if the server is strictly necessary. Having read about hole punching, I am under the impression that the purpose of the server is to allow the clients to exchange some information (including but not limited to their addresses and ports they want to communicate on) so that they can proceed to talk directly. Assuming the clients already had each other's information (again, including but not limited to their addresses and ports), would the server still be necessary? If so why and if not, how could this be implemented?
For instance, say one were to build an application where client_A prints out all information that would have been transmitted to the server for user_A to read, who then sends this to user_B, who then submits this info to client_B (this could be done via email for example). Wouldn't this avoid the need for a server?
Here is another explanation of why I think it might be possible to remove the server in the middle:
In NAT hole punching (assuming I understand it correctly), the communications begin when client_A sends a message to the server. The message contains some information that the server then passes on to client_B when client_B contacts the server. After this point, client_A and client_B are able to communicate directly without the need for the server. I am under the impression that once a direct connection between client_A and client_B has been established, the server could go offline and the two clients would still be able to communicate directly with one another. If this is the case, then I would imagine that any information that is being used to maintain this connection (be that addresses, ports, or any other kind of info) could be exchanged through any other channel (eg: email, a handwritten letter, a voice call, etc) at the beginning of the protocol, and then the connection could be established without ever needing the server.
Regarding 'tricking' the router
As manishig pointed out to me in a comment (thanks), NAT hole punching also requires tricking the router. If I understand correctly (please correct me if not) the router is tricked by having the router store the info for directing incoming packets from the server to client_A, however, these packets are actually coming from client_B after the initial phase of the protocol. If this is a correct description of the problem, is there a way to trick the router that doesn't require using a server?
There are ways to communicate between two remote computers over the internet without an intermidiate server, but IMO it is not the preferred way.
Why an intermidiate server is needed?
If client_A and client_B are both in the same LAN (e.g your home/office network) you can make sure (configure on the clients side and/or the router) that they will have a static ip address over this LAN and they can just talk freely. E.G: If client_A is listening on port 8080, client_B can create a connection to client_A_ip on port 8080
Over the internet any packet sent is passed through NAT usually at least twice. One time after going through your LAN (e.g your home/office router) and at least once over an ISP endpoint. Which means you have no controll over the public ip and port assigned to your packet.
Now not only that you don't have controll over your packet's assigned public ip and port, these are also not static. They won't change while you have an active TCP connection, but you don't have any other guarantee from your ISP regarding your assigned public ip and port.
The intermediate server`s purpose is to dynamically update each client with it's peer info and also keeping the tcp connection open, so that peer to peer comunication will be available.
Alternative solution to an intermidiate server (Not recommended)
If you want your clients to communicate without an intermidiate server you can buy a public static ip from your ISP (if they support it) and then there are ways you can make (with some config) that one of your clients have a public static ip and port that the other client can connect to.
But I wouldn't recommend it, since it requires some understanding in IT and security risks.
Also if both client's are portable and connect to different networks all the time it's not a valid solution