[Nml-wg] Labels (was: Notes from NML call on May 3)

[Freek Dijkstra]

Jeroen van der Ham wrote:

> - Label proposal from Roman on the list describes the outcome of the discussion last week.
> Freek notes that we should only allow one label per port.

In the call I mentioned one exception: source and destination address.

But I have some good hope that this is not an issue. In G.800 terms, 
this seems a different thing. What we're talking about is a "resource 
label", used in multiplexing adaptation and channels in links. "Source 
and destination labels" are apparently used for termination, and thus 
something else (for those interested: Section 6.7.1.1 of G.800, but also 
section 7 and 8 -- in particular Figure 18).

Also, I was briefly worried that VRFs would complicate things when 
describing IP addresses as labels. Historically, routers routed based on 
destination IP address only. Most modern routers support VRFs, which -I 
thought- routed based on source and destination IP address. Turns out I 
was wrong: they route based on destination IP address and source *port*, 
which -I think- can be described in current NML using multiple Switching 
Services.

> This makes it possible to described configured networks very well.

The motivation for the one-label limitation is mostly simplicity.
Multiple labels might be possible, but complicates things (e.g. do we 
need to specify order?; can there be a stack of labels, isn't that the 
same as adaptations, and if so-how to related which label to which 
layer, etc. I rather avoid these questions for now). If this is too 
strict, we can extend things later.

Regards,
Freek