Re: IP namespace (Was: Re: root knowledge)

[Christian Huitema <[e-mail]>]

Hello,

I see that you are investigating various ways to represent the IP address space in
the DNS. Let me tell you of three possible criterias:

1) It shall be directly based on the address itself, not on a DNS representation
of the address,

2) It shall be distributed. Storing tens of millions of entries in a single space
is not an alternative.

3) The distribution shall be independant of the address format. Class A, B and C
are used now, but a new "CIDR" format is just being defined for coping with the
"address exhaustion" problem.

This is in fact a case where you would like to use "longest match" searches, i.e.
select all possible entries in the database where attribute X contains the "first
bits" (yes, I said BIT, not octet) of the target Y, then select the "longest X"
and continue the search. Similar searches would be absolutely needed for NSAP
resolution.

There is no "longest match" operation defined in X.500 -- too bad! We could
however define a particular syntax (subtype of bit string, or of octet string),
and specify that the "<" and ">" operation refer to the partial order:

  X < Y  <=> (length of X, lx =< length of Y) and (X == firsts "lx" bits of Y).
  X > Y  <=>  Y < X

Playing indexing games with this would be relatively easy...

Christian Huitema