Re: GMPLS issues (was - GMPLS Last Calls)

[Yakov Rekhter <yakov@juniper.net>]

Neil,

> Eric, I understand your points on the 'management network'......that is, the
> addressing of it's user-plane is a separate network to the user-plane
> offered for cutsomer connections.  So let me concentrate on the 'external'
> user-plane here in an attempt to answer your question.
> 
> Seems we have some choices in GMPLS at the moment:
> Addressing - v4 or v6
> Signalling - RSVP or CR-LDP
> Routing - OSPF or IS-IS (but exterior is BGP only)
> 
> All these facets are, in principle, independent choices.  And, in
> particular, since 'server layer trails = client layer links' then addressing
> cannot be from the same space at a server and client layer.  The addressing
> might have exactly the same format, but it comes from a different space, ie
> address X at layer M is not = address X at layer M.
> 
> Given that the layers have to be decoupled in the general case (for
> commercial as well as technical reasons - see the BT 'Freeland' requirements
> draft to 12/00 mtg) then there seems to be no reason why an operator could
> not use NSAPs for the access points of SDH or OTNs if he so wished.  

Please note that there is a way to carry NSAPs in IPv6 addresses,
although it does not support all types of NSAPs.  (see rfc1888 for
more details). Please also note that GMPLS supports IPv6 (as well
as IPv4). Finally, please note that a "perfect" solution (the one
that addresses all the requirements of everyone) is a *non* goal
for GMPLS.

> And,
> given this choice, that operator might also prefer to use PNNI as the
> signalling/routing solution too.  So, the issue seems not one of
> 'necesssity' but simply choice......though there may be certain technical
> benefits of choosing 1 solution over another.  I don't believe there is
> anything technically wrong with saying this and giving operators/vendors a
> choice here is there?.....if there is please point out the problems.

Few points:

1. Providing multiple choices for performing essentially the same
thing, especially if these choices have similar characteristics,
doesn't help to develop multi-vendor interoperable solutions.

2. The fact that in the IETF there are not one, but two link-state
IGPs (OSPF and ISIS) is a bug, and not a feature. 

3. The same applies for MPLS signalling (CR-LDP vs RSVP).

4. Suggesting yet another routing/signalling protocol, in addition
to GMPLS, is adding yet another bug (not a feature).

5. I understand that BT submitted a contribution to the ITU
suggesting PNNI as (yet another) routing/signalling for optical 
cross connects.

Yakov.