[Nsi-wg] Groups and Labels

[Jeroen van der Ham]

Hello,

Yesterday in the call Jerry and I briefly touched on the issue of Groups and Labels in network topologies.
After numerous discussions, it seems that Jerry and I still have opposing views, and we do not seem to be getting closer to a solution. This is why I'm raising this as a group issue, so that we can have a broader discussion on this.

The issue at stake here is handling network connections with labels.
Take for example a circuit from CERN, through USLHCNet, NLR, ending up in San Antonio. One end of the connection uses VLAN 3000, this is not available end-to-end, so it gets converted to VLAN 3067 along the way.

My viewpoint is that this label is an essential part in the definition of an end-point of a connection. The end-point is defined both by the place in the network (URN), as well as the label VLAN 3067. Both these elements, point and label, must be indicated, and available to the pathfinder in order to optimally calculate a connection. This does not pose any restrictions on how that calculation and/or negotiation takes place, but at least that information is available.

From what I understand, Jerry's viewpoint is that the label is *not* an essential part of the end-point definition. An end-point is in the network, and a Group of STPs is described that derive from that end-point, which implicitly encode the label through the mapping. The value of the label is available to the intra-domain manager, but not inter-domain. Inter-domain path calculation takes place without knowledge of the label, assignment of labels is done by the domains themselves.

My problem with the latter approach is that it is virtually impossible to do pathfinding when you mix domains that can and cannot swap labels. In the example above the label is translated by a transit-domain. If the label is not available in the pathfinding process, the domains basically have to guess at which label to use. If it turns out not to work, the requesting agent is groping in the dark for another possibility.

Almost all circuits that are being made currently use a label. The OSCARS interface supports the user defining a label in a request. OpenDRAC supports the user defining a label in a request. Other NRMs probably do also, so why would NSI not support this?

Jeroen.