IDs: summary of NVO telecon discussion

[Ray Plante]

Hi IVOA-ers,

You might recall the email discussion we've been IDs.  Issues discussed 
included requirements for IDs, what it means for two instances of an 
ID to be identical, and when should an ID be reused when an object is 
replicated within the VO grid.  An overarching theme of the discussion is 
the scope of an ID framework: what are IDs meant to accomplish?

In an attempt to clarify the scope, the NVO MWG took up this topic in its 
telecon on Feb. 6th.  The goal was to come to some consensus (within the 
NVO project) about what IDs are meant to accomplish, and use this 
consensus to pare down a set of proposed requirements.  The purpose of 
this message is to share the results of this discussion with the wider 
community.  Your comments are welcome.

The minutes of the meeting are accessible via 
http://archives.us-vo.org/metadata/0597.html.  The main points we were 
able to agree on are as follows:

  1.  The primary goal of an ID is to allow retrieval of a unique 
      description of an object or concept given its ID.
	*  while a particular ID may reflect ownership, containment, 
	   and/or derivation, the system need not depend on it.
        *  in general, relationships between entities are determined 
           definitively via an examination of the metadata. 

  2.  The ID framework should maximize freedom of data providers to
      choose identifiers.
        * Providers should control what their various IDs refer to 
          (resources, missions, datasets, services, etc.).
        * Providers should control a namespace that can be used to turn 
          local identifiers into global ones.
        * The framework should minimize the requirements on the form of a 
          local identifier. 

  3.  Services specify a use of an identifier should be as specific as 
      possible as to what the identifier should refer to (e.g. an image,
      a version of an image, the collection an image derives from, etc.)

These assertions were folded into our (now reduced) set of proposed 
requirements on IDs which is available via 
http://rai.ncsa.uiuc.edu/~rplante/VO/metadata/oidreq.txt.

We certainly don't consider the above assertions, nor the resulting 
requirements, to be the final say on the topic.  In particular, there are 
some issues regarding handling persistant names that could use some work.  
I expect that our understanding of the role of identifiers will change a 
bit as we play with identifier specifications and experiment with 
prototype registries.  

cheers,
Ray