On Nov 17, 2011, at 1:01 AM, Andrew Sullivan wrote: > On Thu, Nov 17, 2011 at 02:47:45AM +0000, [email protected] wrote: > >> The DNS server selection document indeed assumes applications do not >> provide preference on what "context" they want to use - because >> generally applications just don't know that information, and we want >> to avoid bothering users with network selection UI dialogs. > > No, the reason applications don't have that interface is because we > told them that the DNS is a unified name space. That's its design. > (This is part of Olafur's point, if I read him correctly.) > > As a matter of fact, because of what people have done with the DNS, it > is _not_ a unified name space, and you will indeed get different > answers on different interfaces.
In other words, people have used DNS in a way that is contrary to its design, and in the past it has happened to (often) work only because hosts tended to have only one active interface to the network that provided DNS at any given time. > Moreover, vendors (who perhaps ought > to have known better) have attempted to fix that by adding curious > different ways of not experiencing too much pain whenever this > situation arises. > > The reason that we need the server selection document, therefore, is > that we want people to do this hackery in a way that causes the least > harm. If we want that to happen then the server selection document should make it explicit that the results of any DNS query via any network interface MAY be used to establish a connection on any network interface. Or maybe there needs to be some way to distinguish name servers serving private namespaces from those serving public name spaces, so that the host can tell the difference. Ideally that should be communicated by the private name servers themselves in every DNS response that involves a private name. Then the host would know that that specific response was interface-specific. Keith _______________________________________________ mif mailing list [email protected] https://www.ietf.org/mailman/listinfo/mif
