On Sep 8, 2009, at 12:26 PM, Jerry Sobieski wrote:
John Vollbrecht wrote:
comment below On Sep 7, 2009, at 5:45 AM, Guy Roberts wrote:
If path computation is the only proposed use for transitional links then I am not sure that we need to use the concept in NML, or are we proposing another use that I have missed?
I think that transitional link is a way of flattening topology where it is possible to do adaptations between layers. It requires that a device be represented at both layers and have a link between layers with a specific adaptation. When doing pathfinding using such links one must be sure adaptations and deadaptations match such that the client info is passed through.
I think this concept is highly useful for describing a "routing area" over which pathfinding can be done. We should consider this relative to other methods of pathfinding which seem more difficult to use in practice.
John
Hmmm...I guess I am confused. I thought transitional links *were* adaptation components of the topology (and vice versa). A selected path that transited/transitioned an adaptation componet had to configure that adaptation at provisioning time; I think though, whatever you call it or where ever it is in the topology, transitional links / adaptation components function differently in two different situations: a) Encapsulation, and b) stitching. The former, is a "vertical" transition where the upper layer protocol is tunneled in its entirety through the lower layer protocol (ala IP/ Ethernet, or Ethernet/sonet (via GFP adaptation) ) and must have a matching decapsulation function at the egress, and the latter is more "horizontal" transition where the current transport protocol is stripped in its entirety leaving only the user data payload which is then placed in the next transport protocol for forwarding (the stitching adaptation does not require a matching function at its egress point - only whatever it needs for the next stage). Does this jive with the discussion and other papers on these concepts?
Probably we are both confused. Jeroen and I have had a long conversation about transitional links and I think he and I agree (at least temporarily) on how they might work between layers. I think your distinction between a case where client data is encapsulated in one form (e.g. VLAN) on one layer and in a different form on another (e.g. SONET). In one method the VLAN is encapsulated as ethernet in GFP and carried in SONET. In this case the encapsulation and decapsulation coming in and out of SONET need to match. In the second method the client data is encapsulated in VLAN, decapsulated between layers and re-encapsulated in SONET. When in leaves SONET it might go directly to a user, or might be encapsulated in a different layer. There does not need to be matching encapsulation and decapsulation, everything get converted back to basic Client Information then encapsulated in a different layer. Are these your distinctions or is it something else. If so - I don' t think it matches the papers, but it is a very interesting concept :) John
Jerry