> > without a mechanism to map the endpoint identifier to an IP address, > > such identifiers are useless in referrals between application > > components. > > This is not so. Read again what I said before: > > If you construct the protocol interactions such that you don't *need* to > be able to look up the "identity->address" mapping (which is what HIP > does - in general, by providing the identity->address mapping used in any > given transaction as part of the initiation thereof), there's no problem.
that's like saying that if I write my protocols so that I never use referrals, then there's no problem with the inability to do referrals. of course, writing my protocols so that I never use referrals drastically limits what I can do with the network. for instance, DNS could not exist. but never mind the lost functionality, we can at least pretend we have a clean model. > So if I have a system which doesn't provide a directory of mappings from > endpoint identifier to addresses, then in the case you cite, when I refer > one application component to another, all I need to do is either: > > - i) provide some other name, one that can be mapped into both identifier > and address, or thus reducing the problem to a different unsolved problem. > - ii) pass the other party both the identifier and a current, working > address for that endpoint. thus requiring me to continue to use IP addresses in referrals. Keith
