[URN] NA id request - wanting a specific number

[<[e-mail]>]

John A. Kunze writes:
 > I'm setting up a prototype URN resolution service for the US National
 > Library of Medicine (NLM).  First off we need a Namespace ID (NID).
 > 
 > The NID that I'm requesting is the string
 > 
 > 	"1000"

Nice number.  Who gets "2000" or other great numbers: "0".."20"?  

My point being that if all you have to label things with is numbers,
then people will assign meaning to the numbers.  Why did you pick
"1000"?  People assign meaning to numbers even if they don't really have
to, though it is relatively rare.  (A couple examples: the TV show named
after a zip code, the Boston cafe named after the original street
address.) So numeric ids do not really avoid the problem of semantic
drift - they merely push it off to another level.

On the other hand, numeric ids are harder to remember, harder to assign
meaning to, and people will be more inclined to use identifiers that are
easier to deal with.  I understood that people who are arguing for
numeric ids are also assuming there will be a layer of human-friendly
ids on top of them, but I have never heard a proposal (or half-baked
idea) for how that higher-level layer of ids will work, and more
importantly, how they will avoid the same problems that other
semantically rich ids have.

The system of DNS names and IP numbers is a good example to learn from,
actually.  You all know how it works, but I'm not sure we all learned
the same thing.  The IP number of a machine can be changed, or inversely
the machine that an IP number is associated with can change.  IP numbers
are rarely used by humans, however, and the DNS system was set up to map
to IP numbers.  DNS names can change for the same IP number, but more
typically, multiple DNS names map to the same IP number and the same DNS
name may map to a different IP number over its lifetime.  So DNS names
are very much like the names of resources, but the resources in this
case are whole machines.  The DNS system supports very scalable
resolution, primarily because it is hiearchical, but the management of
DNS name "assignment" has never properly delt with the issue of
conflicts.

By the way, perhaps a hidden assumption behind the URN work is that
resources smaller than whole machines (e.g. documents and services) need
to be separated from the whole machines.  It is not that the identifiers
of whole machines cannot be "location-independent" names - DNS names are
exactly that.  More generally, resources need to be independent of the
*collections* they are contained within since they may be eventually
removed from those collections.  (A machine or host is a sort of
collection too.)  Perhaps this is why people object to any hierarchical
name system.  But creating a set of naming authorities each with its own
flat name space does not really solve this problem since each flat name
space is still a collection that the resources need to be independent
of; eventually some of the naming authorities will fail.  The only
solution that avoids collections altogether is to have only one flat
name space for all resources.  But then we have a serious problem
achieving scalable resolution.

 > As an aside, in discussing
 > 
 > 	http://www.ietf.org/internet-drafts/draft-ietf-urn-nid-req-03.txt
 > 
 > with Leslie, she had the idea that when the requestor doesn't specify a
 > NID, that the number assigned by the IANA-like body should be prefixed, say
 > 
 > 	iana-NNNNNN

I believe the only way that numeric ids could work to avoid conflicts is
to assign them.  Better avoid re-assigning numbers too, but that should
not be a problem as long as the number of digits can grow indefinitely.
But then, alphanumeric ids could be assigned too.  Consider license
plate "numbers".  Then consider vanity plates.

So I believe the issue of numeric ids is a red herring.  The real issue
is whether ids are assigned, or if they are chosen, how conflicts are
resolved. The proper way to deal with top-level name space conflict
issues is to inherit another system for dealing with such conflicts,
such as the trademark system.  Or if you want to set up your own system
for dealing with conflicts, it better be at least as good.

--
Daniel LaLiberte
 [e-mail]  (was: [e-mail])
 [e-mail]