Keep in mind that this paper was outdated the moment it was published and announced on p2p-hackers mailing list [1]. It doesn't deal with the "long tail" of NA

Keep in mind that this paper was outdated the moment it was published and announced on p2p-hackers mailing list [1]. It doesn't deal with the "long tail" of NAT devices that increment/decrement ports in (somewhat) predictable manner and _most importantly_ it describes hole punching as a client-driven process. The latter is the crucial point. By tasking a dedicated (mediating) server with coordinating the punching sequence it becomes possible to time the process much more precisely and to help predictions to actually match the reality. Combined with a bit smarter port predication it brings the success rate from 80% to 95-97%... or at least it did 10 years ago when I was using this for Hamachi P2P VPN, though I suspect that very little has changed in terms of NAT type distribution since then. [1] http://copilotco.com/mail-archives/p2p-hackers.2005/msg00126.html http://copilotco.com/mail-archives/p2p-hackers.2005/msg00126...