Re: radical solutions

[smd@ebone.net (Sean Doran)]

| There are some interesting details buried in here. If you want to map
| from the lower 8 bytes into something that can be used in the upper 8
| bytes for routing, one likely needs a heirarchy in the lower bytes
| (e.g., ala DNS delegations)

riiight, but your distributed hierarchical database can have several
interesting properties, such as thaler/ravishankar top down dynamic
hierarchicalization, or even registration into the database by entities
not hosting the database.

with eui-64s, it would be straightforward to carve things up on
byte boundaries and provide a mechanism for registration of an
NS RR or suchlike into the ultimately correct zone.   NS RRs are
used in RFC 2317 which deals with divergence between database distribution
hierarchy and registrant hierarchy, although if the only thing one
is looking up in the DNS-like database is a "PX" record to stuff
into the top 8 bytes, 

the real problem is undistributing mappings to avoid lookup delays
and traffic for traffic arriving at a gateway that changes the RG
really does require a hierarchical structure (or LOTS of memory).
some of the work done on self-organizing caching schemes may prove
useful to allow for a partial centralization of copies of the data 
"near enough" topologically to make the cost of a lookup nonfatal,
although i really have my doubts.

would it be possible to use the multiple-address-per-host feature
of v6 to bootstrap from eui-64 autoconfiguration address to 
something that changes infrequently but which is hierarchical?
that makes the lookup on a key in a large, sparsely-populated
distributed database happen perhaps as infrequently as when
booting any given machine (i.e., the machine gets enough connectivity
based on its eui-64 address to get a neatly hierarchical identifier
which a router far away can use as a key to get the top 8 bytes in
reasonable time).  indirection being the better part of valour and all that.

	Sean.