Hi all, Last year we have had a few discussions on identifiers in the NML-WG, as well as in the GLIF. There have been two more or less conflicting solutions. Nevertheless, I hope to reach consensus on a way forward at the coming OGF meeting, which coincides with the GLIF meeting. Let me try to summarise the two points of view: 1) The theoretical view: an identifier is simply a unique string. No information at all should be implied in it. "sdimfsdhdsjkjd39n34n58" is great, even more if there is a good method to make it unique. E.g. "mydomain.net:sdimfsdhdsjkjd39n34n58". The great thing is that you can change the properties of the thing without changing the identifier. 2) The practical view: we should re-use the NURN (network identifiers) as used in the perfSONAR software suite, e.g. "urn:ogf:network:domain=mydomain.net:device=mydevice:port=4:link=1". This conveniently has all the attributes inside, very easy to parse. No need to make it more complex. I have been an advocate of the first point of view, Martin has been a vocal advocate of the second view. Of course, both view have their merit. Jeff proposed to make the order of items in the perfSONAR identifier fixed, so at least a string-wise comparison is possible between two identifiers. Freek said he could live if the first part of the identifier implies the originating domain. I wonder if we can make progress by reaching consensus on the following statements: - The NURN is not an identifier in the formal sense, it is a couple of properties that together uniquely identify something. - That said, sometimes it may be overkill to create a special identifier, but it is good enough to identify something by it's properties, as the NURN does. (and perhaps more controversial:) - The identifier that is sent to external parties consists of a domain part, identifying the originating domain, and a unique string part. Whilst this string may contain more properties, other parties do not need to interpret the meaning of this string. - As the NURN is formally defined in the OGF (I presume it is in the NM-WG), if the NML-WG deviates from this format, the NML-WG has an obligation to also define a practical transition strategy which does not suddenly break things. Regards, Freek
On Feb 12, 2009, at 3:58 AM, Freek Dijkstra wrote:
I wonder if we can make progress by reaching consensus on the following statements: - The NURN is not an identifier in the formal sense, it is a couple of properties that together uniquely identify something. - That said, sometimes it may be overkill to create a special identifier, but it is good enough to identify something by it's properties, as the NURN does. (and perhaps more controversial:) - The identifier that is sent to external parties consists of a domain part, identifying the originating domain, and a unique string part. Whilst this string may contain more properties, other parties do not need to interpret the meaning of this string. - As the NURN is formally defined in the OGF (I presume it is in the NM-WG), if the NML-WG deviates from this format, the NML-WG has an obligation to also define a practical transition strategy which does not suddenly break things.
The big folks are the DICE ones since they're using the NURNs already, but the plan is for them to adopt what the NML/NSI groups decide upon, so we should try and get this right before much further adoption occurs. The GLIF folks have been discussing identifiers for circuits and seem to have reached a consense on ones of the form "[domain.edu]: [opaque goo]". The plan is for them to eventually move to a URN approach of "urn:glif:domain.edu:[opaque goo]" or similar. I think in the interest of cooperation with the other groups, and the desire to have one unified identifier scheme, it's probably best to match that style closely. Something along the lines of "urn:nml:domain.edu: [opaque goo]". As long as there is a URN header so that we can some idea on how to grok a given identifier, I'd be fine with that scheme. Cheers, Aaron
On Feb 12, 2009, at 9:48 AM, Aaron Brown wrote:
On Feb 12, 2009, at 3:58 AM, Freek Dijkstra wrote:
I wonder if we can make progress by reaching consensus on the following statements: - The NURN is not an identifier in the formal sense, it is a couple of properties that together uniquely identify something. - That said, sometimes it may be overkill to create a special identifier, but it is good enough to identify something by it's properties, as the NURN does. (and perhaps more controversial:) - The identifier that is sent to external parties consists of a domain part, identifying the originating domain, and a unique string part. Whilst this string may contain more properties, other parties do not need to interpret the meaning of this string. - As the NURN is formally defined in the OGF (I presume it is in the NM-WG), if the NML-WG deviates from this format, the NML-WG has an obligation to also define a practical transition strategy which does not suddenly break things.
The big folks are the DICE ones since they're using the NURNs already, but the plan is for them to adopt what the NML/NSI groups decide upon, so we should try and get this right before much further adoption occurs. The GLIF folks have been discussing identifiers for circuits and seem to have reached a consense on ones of the form "[domain.edu]: [opaque goo]". The plan is for them to eventually move to a URN approach of "urn:glif:domain.edu:[opaque goo]" or similar. I think in the interest of cooperation with the other groups, and the desire to have one unified identifier scheme, it's probably best to match that style closely. Something along the lines of "urn:nml:domain.edu: [opaque goo]". As long as there is a URN header so that we can some idea on how to grok a given identifier, I'd be fine with that scheme.
I like this plan. jeff
participants (3)
-
Aaron Brown
-
Freek Dijkstra
-
Jeff W.Boote