Re: draft-ops-endpoint-mib-08.txt

[Harald Tveit Alvestrand <Harald@Alvestrand.no>]

At 13:48 07.06.2000 -0400, Thomas D. Nadeau wrote:
>         That is a good point. What I can say then is that ONLY v4 and
>v6 IP addresses are allowed.
>
>>Again, why does a DNS name not work?
>
>         Because MPLS/RSVP does not use DNS names to
>DNS names to signal MPLS tunnels.

[just sticking my head in]
Note that a syntax definition can never capture ALL the semantic 
restrictions on objects.
In this case, it's unrealistic to expect the MIB syntax to catch the idea 
that both ends of an MPLS tunnel use the same IP address family, or that 
it's unrealistic to expect to use 127.0.0.1 (localhost) as one end of a tunnel.
Or, since global uniqueness is required, 10.x.x.x addresses.

Given requirements for IPv6 rapid renumbering, it's not even completely 
unthinkable that a future version of MPLS will use GUIDs or DNS names for 
unique tunnel identifiers (I assume you're replacing MplsTunnelCookie from 
draft -05 with something that has more strict syntax?)

I'd recommend staying with comment, not syntax, saying that this value 
reflects incoming data from an MPLS label, and only values that are 
permitted there can be used here.

(note - using the global label as a tertiary index will not necessarily 
make it simple to find an entry given that you know the MPLS label. If 
that's what you want, it's better, IMHO, to keep the strictly local 
indexing scheme and add a new table giving mapping from globally unique 
MPLS label to tunnel table index. But since I don't fully understand draft 
-05, and haven't seen the forthcoming version, I may be off base here.)

                       Harald

--
Harald Tveit Alvestrand, EDB Maxware, Norway
Harald.Alvestrand@edb.maxware.no