Hallam-Baker, Phillip wrote: > It is quite easy to see how an application that is designed to tolerate > renumbering is able to cope with it given appropriate O/S and protocol > level support. I suspect what is happening there is that SSH loses the > connection and then transparently attempts to reconnect before telling > the user that it has failed and dropping the entire connection state. > > But most IP applications are not designed to maintain connections for > days, SSH is a rarity in that respect. SSH is also a rarity in that there is usually a good correspondence between a DNS name and the specific entity to which one wishes to connect via SSH. So it makes more sense for an SSH client to try to reconnect to a peer and continue a session than it does, say, for an FTP client. > Renumbering your network every day is probably quite practical. I seriously doubt that. There are very few layer 7 protocols which can gracefully tolerate that kind of disruption, and our architecture doesn't really give them a good way to do that. And it's not just a naming problem. Granted that if we were to start renumbering more frequently, we'd get better at it. But the Internet architecture as it currently exists doesn't come close to supporting that. Which of course, is part of why people continue to want NAT - so at least they can avoid renumbering for local applications. Keith _______________________________________________ Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf