Guy Roberts wrote:
As I see it, there are broadly two ways of approaching pathfinding in multi-domain, multi-layer networks:
Approach 1: AutoBAHN like - the layers are collapsed into a single abstracted layer and pathfinding is done on this layer. We then perform stitching on a set of possible paths.
Approach 2: path finding is done on a complete multi-layer graph with full knowledge of layer adaptations. A much more limited (if any) stitching function is then required. I think this is more like the method proposed by Freek in his thesis.
Guy, your analysis is spot on. Thank you for your insight. The notes on "Conversation about ITU concepts with Ciena folks" contained:
There were some concepts indicating that GMPLS was not as good at describing multi-layer capabilities as G.800 (and possibly NML). I don't really understand precisely why.
John Vollbrecht added (off-list):
I would love to have you or someone make a more specific/detailed critique of GMPLS. Is it just for layers that it has problems?
It is just the fact that GMPLS does not represent the relation between links on different layers explicitly. This give problems in either of these situations: - incompatible adaptation functions (e.g. GFP-F vs. LEX or Q-in-Q vs. PBB-TE) - multiplexing and inverse multiplexing Basically: it gives problems whenever the collapse as describe in approach #1 above leads to loss of information in the resulting graph (indeed, I proved that there are -rare- network topologies where any possible collapse must lead to loss of information.) Regards, Freek