>> And you talked about Stuart Cheshire described a couple of IETFs ago, >> Could you help to point out the link? > > Sadly, I don't have it, but I suspect Stuart does, and I'm pretty sure he's > reading this. Thanks, let's see whether he is going to talk here. > The gist of what he was saying is that if you have an IPv4 address that > looks okay, and an IPv6 address that looks okay, you can't assume that they > *are* okay, because there may be no route to the global internet on either > your IPv4 or your IPv6 link. So you must attempt to use both addresses, > not just one, and you must do it at the same time. Whichever one answers > first, you take. It means that the new IP model will try host's multiple connections each time as you suggested as well. > If you prefer either IPv4 or IPv6, and the transport you preferred happens > to be the one that was broken, a smart user will disable the one you've > preferred. That user will then advise his or her friends, for example, > that "IPv6 creates instability, so you should disable it." This impedes > deployment. > > The unattended multiple interface situation is quite similar. I think the > attended case (a laptop with two or more network interfaces) is actually > better handled through user intervention, because the user has knowledge of > the physical situation that would be difficult to communicate to the > computer. But in the unattended case, you can get into the same sort of > "wrong learning" situation, where a smart but naive user who debugs a > network problem winds up learning a workaround that would impede > interoperability if everybody did it. Interesting thoughing, workaround will impede deployment. Thanks -Hui _______________________________________________ Ietf@xxxxxxxx https://www.ietf.org/mailman/listinfo/ietf