October 29, 2006


I am very fond of the Anderson Valley, California.

The largest town is Boonville.  It's a beautiful place.  (And there are several very good wineries in the valley.)

They have a language there, Boontling, that was indended to be incompressible to outsiders.

I just read Vint Cerf's planned opening for the IGF

It would deny to the users of the internet their right to create their own Boontling.  Indeed the denial of such dialects in the domain name space has been one of the goals of ICANN policy.

This is misdirected and assigns attributes to domain names that simply do not exist.

I dealt with these widespread, and oft repeated (as Vint does) errors in my submissions to the US National Research Council.  For convenience I will restate them here.

Vint alludes to a concept that goes under the name "Global Uniform Internet Name Space".  Globally Uniform Internet Names are known as GUINs.

Are domain names and the domain name system powerful enough to provide GUINs?  Vint's talk seems to assume that they are.  My analysis, below, indicates that they are not.

The properties of a GUIN are these:

  • Universal validity or universal invalidity
  • Invariance:
    • Location invariance
    • Client invariance
    • Temporal invariance

Let's look at these in a bit more detail:

Universal Validity or Universal Invalidity requires:

  • Every name that is valid must be valid everywhere.
  • Every name that is invalid must be invalid everywhere.

Location invariance requires that every valid name must have the same meaning no matter where it may be uttered.

Client Invariance requires that every valid name must have the same meaning no matter by whom it may be uttered

Temporal Invariance requires that once a name becomes valid it must have the same meaning no matter when it may subsequently be uttered.

Under these requirements, domain names are inadequate to operate as GUINs.

Due to caching and the time to propagate changes domain names are not strictly universally valid, but they are not that far from meeting the criteria.  So we can fudge a bit and say that domain names meet the requirement of universal validity or invalidity.  (It is this property of validity/invalidity that was damaged by Verisign's "SiteFinder" proposals.)

Where domain names fall down badly is on the requirement of invariance.

Failure to meet client invariance: an example: Amazon.com uses returns different web content to different users even though they each provide the exact same URL containing the exact same domain name.

Failure to meet location invariance, an example: The domain name "www.google.com" when uttered via HTTP or HTTPs will result in different web pages depending on where the query comes from (e.g. if you are in Sweden you will be received the web page normally visible under "www.google.se" even though the user uttered "www.google.com".

Failure to meet temporal invariance: some examples. The IETF's "internet drafts" are visible via HTTP for a period of time, after which their content disappears.  We often call this URL rot.  Domain names are quickly recycled, one day they may result in useful content, the next they may have been transferred to a new operator and be filled with "domain monetization" advertisements.

The point of this is that Vint's talk, and that of others, is presuming too much of the domain name system.  They are assigning attributes that it simply does not have.

From the point of view of internet Governance the point to be taken is that we ought not to try to build a system of governance based on an incomplete appreciation of what a technology such as DNS does and does not do.

And finally: What is wrong with local dialects anyway?  Vint's presentation presumes that everyone has a right to reach out and communicate with everyone else.  Why?  Don't people have the right to chose to be unreachable and to interact only with their chosen group?

In other words, what is wrong with internet Boontling?

Update: I am being a bit imprecise about the difference between a domain name, a URL, and how these are used by various protocols, such as HTTP.  But that does not diminish the force of my argument that domain names are being shopped around as having properties that they simply do not have.

Posted by karl at October 29, 2006 9:33 PM