There is a problem with the experimental scenarios which use the DNS for resolving URNs - e.g. uri.lut.ac.uk. Domain names are actually quite variable. They usually incorporate some variant of the organization's name, and often the names of departments and groups within those departments. Unfortunately, organizations and their departments are forever merging, changing their names, and even filing for bankruptcy! This brings us back once more to the dangling links problem.
In an ideal world it would be possible to use abstract names for resources, e.g.
Computing, Unix, Shells, Z Shell
This level of functionality is provided in some distributed directory systems, e.g. in the international standard X.500 system, and in experimental systems such as whois++ and SOLO[31].
X.500 is believed by some to be unsuitable for this sort of large scale usage, because of both the complexity of the underlying OSI architecture, and the lack of a scaleable mechanism for distributing meta-information. However, work is ongoing in the X.500 community to remedy some of its deficiencies and improve integration with the Internet[32,33].
Both SOLO and whois++ use a relatively new technique known as the centroid[34,35]. Centroids explicitly address the issue of distributing meta-information, though the technology itself is in its infancy. The centroid is a collection of attributes and unique values for those attributes, taken from the contents of the server - essentially a lightweight full-text index which can be used to determine whether the server has any relevant information.
Centroids may be generated automatically from the contents of gopher and World-Wide Web servers, and from anonymous FTP archives via IAFA templates. Having been created, it is envisaged that they will be harvested by index servers. Users will be able to find out where the information they require is located by searching these index servers. The index servers themselves would form a sort of widely distributed archie-like service.
In addition to supporting abstract names of the form shown above, centroids make it feasible for users to in effect simultaneously search a large number of Internet servers for information - e.g. with questions such as ``which University library in the UK has the book with ISBN number 0-586-03716-0'' . The search would actually be restricted to a handful of index servers.
Given a place in which to store meta-information about a resource, there is clearly the potential to lodge classification information - subject headings, keywords, and so on. Whether any of the existing classification schemes can effectively be applied to networked information is outside the scope of this document!