.@ Tony Finch – blog


Earlier this week a technical cockup in one of the insane alternate DNS roots spilled onto the NANOG mailing list, to much amusement. Bill Stewart posted some comments on resistance at Bell Labs to the deployment DNS in the mid 1980s, including “The Hideous Name” by Rob Pike and P.J. Weinberger, linked above.

It’s very wrong, and it’s interesting to understand why because it provides a historical perspective on the state of networking at that time. The paper seems to have an underlying assumption that balkanization is the natural state of networking. This has two corollaries: that you do not need to (and should not) compromise your user interface to accommodate the naming schemes of other networks; and that a global consensus name space is not possible.

The paper’s model of a name in computing is of a path, which is intimately related to the network’s topology - a topic which is of no interest to people who aren’t network engineers. In this model the idea of a relative name makes sense: how do I get there from here. However they massively underestimeate the value of globally-valid names, which remain so as they traverse the network and do not need to be rewritten as they move from one namespace into another. (They also massively overestimate the reliability of rewriting.) Globally-useful names are so important that they are created by the users of systems that don’t support them: for example in UUCP, people would quote their addresses relative to a few well-known hubs, which effectively acted as the root of the namespace as well as the core of the network.

This illustrates the network effect, which is an idea that is of course absent from the paper since the paper was written at least ten years before the strength of network effects became obvious. Instead the paper invokes standardization as the solution:

It is clear that standards are necessary for electronic mail to be delivered reliably across network boundaries. What needs to be standardized is the interpretation of names, especially at network boundaries. Until such a standard exists; is syntactically and semantically clean; distributes the interpretation of names across the systems that understand them; and is adhered to, the network mail situation will not improve.
This is of course correct, but the solution we have ended up with does not follow their principles of naming at all - and it is of course centred on the Internet and so can't be used to gateway between other networks directly, which they assumed would be necessary.