Re: I-D ACTION:draft-hendrikx-wallis-urn-nzl-00.txt

Roy Fielding writes:

> Noah has reminded me (off-list) that there is at
> least one other example that has been successful,
> namely UUIDs/GUIDs.

> There are definite advantages to completely
> automated, decentralized naming systems (e.g., DCE
> to COM to urn:uuid:) 

Although the official specs seem to be locked away in ISO/IEC 11578, for 
which there appears to be a charge [1], my impression is that the most 
common means of generating UUIDs/GUIDs is in fact to use Ethernet or 
similar MAC addresses as part of the process.  My recollection is that 
these are in fact handed out in batches on a per-vendor basis, and thus 
are indeed managed centrally.  A Google search yielded [2], which is 
presumably not authoritative but which does support this recollection.

Thus, like the DNS names in http URIs, this common form of UUID does 
indeed trace to a central registry.  The structure of such a UUID is not 
in fact particularly opaque, but the information conveyed says much more 
about the entity doing the generation and the time at which the generation 
is done, than it does anything about the resource identified.   For that 
reason, users rightfully tend to think of UUIDs as more opaque than http 
URIs.  Indeed, one of the great complications in using UUIDs in certain 
applications is the lack of convenient ability to generate them uniquely 
on systems that lack network capability (and thus MAC addresses).  They 
are not pseudo-random 128 bit numbers.



Noah Mendelsohn 
IBM Corporation
One Rogers Street
Cambridge, MA 02142

Received on Saturday, 26 February 2005 03:31:45 UTC