> I'm not sure that bangpaths actually *were* "suitably anarchic" --- if
> "somebody else" had decided to name their own machine decvax or ihnp4
> in the Olden Times, it probably would have resulted in severe
> confusion. Certainly if any intermediate site would have wanted to
> connect to both the "new" ihnp4, and the "old" one (a major uucpnet
> hub at Indian Hills)
I hadn't been considering byzantine failures; I'd been thinking that "of
course" no one would name their machine ihnp4, and no one but those crazy
Soviet Leaders would name their machine mcivax.
> probably even if not, due to attempts at
> optimizing paths at intermediate sites, which made the implicit
> assumption that atomic hostnames were unique.
I am not aware of what happened on the backbone, but out at the leaves where I
was, the path optimization software and databases showed up very late in the
game.
Might uniqueness be recoverable? You'd mentioned using the set of known
neighbors as an implicit root set; in the case of duplicate hostnames, one
could use the set of known neighbors as implicit uniquifiers. Even if there
were two kremvaxen, wouldn't it be reasonable to distinguish decwrl!kremvax
from moscvax!kremvax?
The idea that we could afford to run online antialiasing algorithms was
supposed to be implied in my original question:
> Perhaps now,
> with our better multihop connectivity, we could let the machines
> deal with the grottiness?
-Dave