[idn] A two-level problem?

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

In reading all these messages, I wonder if we're getting too close to the 
problem.
We're discussing DNS modification, character set representation and so on 
as an unit; I wonder if it would be good to pull back a little and see if 
the problem admits of subdivision.

For instance, if we imagine that the DNS service (as contemplated here) 
consists of 2 parts:

- The DNS transport, which is the part that goes from the client's interface
   card to an authoritative server's interface card, including any caches on
   the way, but NOT including what's inside client libraries or master servers
   (indeed also excluding the zone transfer function; more later)

- The DNS service, which is the part that goes from the client application
   program to the information that is published by the zone master.

The DNS transport can, I think, be described in fairly simple terms, such
as:

- Binary-transparent movement of labels and record values; labels are
   divided into pieces of up to 63 octets, and represented in length+value
   format; forget about dot notation - it does not exist at this layer.
- Well known cache matching rules; match on octet values, fold octets
   0x41 to 0x5A to 0x61 to 0x7A for matching by caches, otherwise exact
   match only
- Only records from authoritative servers ever exist anywhere, and not a
   bit gets changed in them (crucial for DNSSEC)

This should be only minor or no mods to the existing DNS service, and gets 
us out of the business of trying to do solutions that depend on 
intermediate-server upgrades (I hope!)

The DNS service layer is where we have the requirements; this is where one
can talk about what *authoritative servers* are expected to match up, such 
as UTR#21 version 1, 2 or 3 or simple case folding or.....

When we're dealing with stuff that is this complex, it's almost certain 
that we have to change it later; if we can split up the model in such a way 
that it's clear that those changes impact *only* the DNS service layer, not 
the transport, then I think we have won a great deal.

And if we can get this conceptual model described in the requirements 
document, I think we can lay to rest a good deal of the worries of some 
people for the operational impact of IDN.

Just a thought and an idea - if it's right, we should be able to worry more 
constructively.

What do you think?

                   Harald

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