[Nml-wg] pathfinding, SF and G.80x

[Freek Dijkstra]

Victor Reijs wrote to the NSI list:

> I think [G.800] is a good standard to look into, indeed. As I and others 
> are looking at the Stitching Framework, it will be interesting to see 
> how this standard maps the SF [Stitching Framework].
> Luckily the ideas are quite similar (it is a unified approach and there 
> is a close 1 to 1 mapping of some the relevant definitions, but there 
> will be some difference of course and we need see how to map these).
[...]
> By the way the Stitching Framework is in G.800 speak: about finding out 
> what LI (Layer Information) makes sure that the Layer processor function 
> can function at an interdomain level ('interdomain' is not defined in 
> G.800 I think, so it is perhaps a specific abstracted 'layer relationship').

Victor,

When looking at explicit multi-layer network descriptions, I have the 
impression that most approaches start with G.800 as the base model (or a 
very similar model) and then extend it in one of these two ways:

- Extend the base model by adding addition properties to the model to 
cope with technology constraints. This is the approach taken by NDL, and 
Stitching Framework.

- Extend the base model by making subclasses for each technology (WDM, 
TDM, Ethernet, IP). This is the approach taken by cNIS, GMPLS, and UNIS.

The advantage of the first approach is that is is truly technology 
independent, which makes it very suitable for inter-domain path finding 
(as the algorithm does not need addition for technology specific, which 
is a problem if new technologies are introduced).

The disadvantage of the first approach is that it is hard to model 
characteristics which are only suitable for a single technology. (Think 
about signal degradation, memory buffers, packet size, which are hard to 
model in a technology independent enough that it applies equally well to 
-let's say- wireless, TDM and Ethernet.)

The advantage of the second approach is that these technology quirks can 
be modeled in a better way using technology-specific extensions. This 
makes this approach very suitable for monitoring or intra-domain path 
finding.

The decision which approach to use will be on the agenda in the upcoming 
time for NML.

I personally hope that it is possible to create some hybrid form, which 
takes the best of both approaches by describing all extensions, no 
matter how hard, in a generic way. I guess the stitching framework is a 
set in that direction, but personally I am not convinced that this 
really works for ALL extensions. I would be thrilled to be proven otherwise.

Regards,
Freek