Re: radical solutions

[smd@ebone.net (Sean Doran)]

Well, Ran got to the message while I was out looking after the world
of v4-derived-revenue today, which takes some of the steam out of a 
desire to make a long reply.

Ran writes:

|         Folks interested in GSE should take the above comments
| as input into how to edit/revise the GSE documentation, IMHO.

I agree.

(However, I was tempted to wonder why the DNS itself (with different
roots & classes), or something DNS-like but with a different,
perhaps Thaler/Ravishankar style, distributed hierarchy, couldn't
form part of the routing system.  Is there any reason why a DNS-like
distributed database couldn't be as dynamic as the current IDR scheme?
We do update that whever a customer <-> isp link fails, give or take
aggregation.

That is, could a router do a "GSE RG encoding.of.8.rightmost.bytes" lookup
and get something to put into the 8 leftmost bytes?   This is vaguely 
similar to RSVP/TE + MPLS labels (don't hit me, Lothberg), except that 
you can prepopulate a local cache to minimize the amount of setup needed for
traffic to destinations that haven't been sent-to in a while.  

I also wondered why hosts would ever look at the first 8 bytes
unless the hosts in questions were also simultaneously routers.
However, as Ran says, these are good questions to answer from
a proper design team, and Erik just added another thing to
Ran's initial list of things that would need examining there 
(viz. DNS on hosts, such that they get AAAA RRs with some "null pattern"
in the leftmost 8 bytes, or something like that)).

| In the absence of current GSE documents to discuss, it seems
| best that we not assume GSE works on this list

Yup, for instance, I personally have no idea what the RG should look like,
or even if or how often its format may change as/if the system evolves.

	Sean.