RE: WG last call: draft-ietf-tewg-diff-te-reqts-04.txt

["Ash, Gerald R (Jerry), ALASO" <gash@att.com>]

Sudhakar,

> 1) How would an LER use the knowledge of BC model in
>    the path setup/compassion?

As one example (there are lots I believe), the LER might use the BC model in conjunction with the LSDB/TED to do constraint based routing of an LSP.

> 2) What do you do if models are different?

If each LSR (or link) used a different BC model I suppose the LER could take that into account in doing its constraint based routing, as in the example above.  However, I don't know how this would be beneficial.

> 3) Can I use different BC models on each link if
>    I choose to rather than one model for the whole LSR?

It seems that you could, see the earlier posting today from Francois.  But I don't know why this would be beneficial.

>    If so "consistency" broken?

Yes, I think so.

> 4) What is the impact of different models in a network?

I suppose it depends on how different the models are.  
For example, if each link in a network has 100 units of bandwidth, and:
- half the links have 1 unit of bandwidth allocated to CT1, 99 units to CT2;
- other half of the links have 99 units of bandwidth allocated to CT1, 1 unit to CT2;
That's pretty incoherent.  
Perhaps you can create an example to show how it is beneficial to use different BC models on the links.

> I have more severe problem with consistency regarding 
> the TE-Class than the BC model. What if each node is 
> configured differently such that one LSR understands 
> a TE-Class differently from another LSR? The whole 
> solution may not work correctly!!

I tend to agree with you here, that's why we wrote an I-D (now expired) proposing standardized CTs (see the presentation at IETF-51 http://ietf.org/proceedings/01dec/slides/tewg-5/index.html).  However, that's an issue separate from the DSTE requirements draft.

Jerry