According to this diagram https://developers.google.com/talk/libjingle/important_concepts#candidates 8% of their connections would have to transfer data through a relay server (but many applications will say 'no' at this point because it is expensive). This study from 2005 (http://nutss.gforge.cis.cornell.edu/pub/imc05-tcpnat.pdf) estimates an 11% failure rate, but when the connection succeeds your STUN server handles only the much smaller amount of traffic needed to establish a direct connection through the NAT devices.
Seems like you might need to know about individual TCP connections at a lower level than you get from ZeroMQ. On Sun, May 6, 2012 at 4:26 PM, Brian Duffy <[email protected]> wrote: > Okay, I read up on Nat traversal (interesting). One more question; is it > worth it? Assuming I am developing an application that would need to handle > port forwarding automatically for users that are not expected to interact > with their routers settings, I am concerned about the drawbacks of Nat > traversal. Specifically, I don't know how many routers might support upnp or > nat-pmp by default, and I don't know if enough routers will be configured in > such a way that STUN will be effective. Also, I really don't want an > external server in my implementation if I can help it. I may just decide to > implement some local blue tooth networking and write a mobile app so that > users can atleast share some data in a personal area network and then sync > back to the clients when in range, but I would be interested in peoples > opinions on implementing automatic port forwarding in their applications and > what luck they may or may not have had. _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
