Phillip Hallam-Baker wrote: >> The problem can be solved by carefully designing connection >> establishment protocols to support multiple addresses of a >> host, which means no solution exists at the connectionless >> layer of IP. >> Modified TCP, which send multiple SYN to several addresses >> of a peer helps a lot to reduce timeout. > I am pretty sure we can fix the problem if we are prepared to adapt > the stack somewhat. FYI, modified socket API and TCP with multiple IPv4 and/or IPv6 (optionally with ID/locator separation) addresses was designed and implemented several years ago. It's not very hard unless you desperately try to solve the problem at the connectionless IP layer. But, I see no point to insist on IPv6 with a lot of wrong design choices. > The alternative is to do nothing and let various people hack the stack > up completely with meat axes and then we will be working round the > consequences for decades. The alternative is to live with IPv4 with port restriction, which is a lot more realistic because we can continue to use the current backbone. > But really, the challenge is that carrier grade NAT works just fine > for the ISPs who have the decision making power here. While legacy NAT is a form of port restricted IP, lack of end to end transparency is still a problem. Masataka Ohta _______________________________________________ Ietf mailing list Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf