On Sep 22, 2008, at 9:58 AM, Jeroen van der Ham wrote:
Hi Aaron,
Aaron Brown wrote:
Martin, Jeroen, Aaron, John, Victor, others: what do you think? For the class definitions themselves, I think it makes sense to use URIs a la namespaces so we could put some documentation at the specified URL.
Just to make this absolutely clear: by "URIs a la namespaces" you mean to use a URL, abbreviated as a namespace? (So "nml:Node" where the namespace nml has been defined previously). It's a little bit confusing because both URNs and URLs are URIs...
Correct, the URL style. mistyped when I wrote it.
For the identifiers for individual instances, I think the URNs make more sense since it doesn't imply a specific method of access to get information about the element.
Now this I do not agree with, because it would mean that the OGF starts administrating its urn:ogf namespace, and handing out specific subsets to domains, with all associated registration and possible squatting problems. Domains already have a domain name, so why not use that?
We do use the domain name. The URN identifiers take the form: urn:ogf:network:domain=glif.is:[domain-specific-chunk] This does minor standardization to allow domain scoping while giving flexibility to the domain to define their own identifiers. The instance identifiers is probably all be out of scope for the NML. The choice of an identifier scheme is really about the lookup and distribution of the topologies. Since we're focused on just describing the topologies themselves, and not specifically concerned about distribution, we can probably just leave it at "use URIs for identifiers", and let the higher-level groups like the like the NMC or the DICE-WG or whoever, decide on the actual structure of the identifiers. Cheers, Aaron