Re: A philosophical question about large directory structures

Graham Klyne wrote:

> This idea is used explicitly in a recently published draft for identifying
> content media feature collections:  an identifier with some randomized (but
> repeatable) property
> that is used to serve as an identifier.  If the identifier is long enough, and
> sufficiently independent of other germane data ("random"), then its uniqueness
> can be asserted with a probability arbitrarily close to 1. (This assumes that
> uniqueness is the reason for central management of the namespace -- there may
> be other considerations.)  (The draft mentioned is
> <draft-ietf-conneg-feature-hash-02.xt>.)

Cool. This is very similar to  the approach to managing negotiation contexts that
was proposed for the original HTTP-NG. Unfortunately, although a checksum of the
content of an object could be used to generate an  identifier,  that content is
mutable, so the ID would change everytime you moved offices.

The rallying cry of whois++ was "there's no such thing as a meaningful
namespace". The purpose of "names" are to uniquely identify an object, and to
provide a fast path for looking up a specific object. Human meaningful work is
done by search - that's where CIP comes in. Once you relax the prettyness
constraint, alllocating names becomes much easier.

There is an interesting application of hashes to the centroid architecture
though. You can apply a hash to a name, split it into 4 chunks, and use those as
index terms, with a bounded dictionary size and a random distribution. This has
some useful properties for SQUID like applications.

There's no real need to standardise schema; applications and organisations will
have different needs at different times. Where standardisation is useful is in
attributeTypes. These can be slotted together as needed.

Simon