forward-last and a BIND question.

[Johan Ihren <johani@autonomica.se>]

Mark,

1. After having thought some more about this, I really do not think
   that a configuration based upon a static, address based
   identification of a translator will work out.

I.e. something a la 

        options {
                ...
                forward-last { 3ffe:507:1ff:2:240:96ff:fe34:bc64; };
        };

   My first objection is that as soon as we start using *addresses*
   under someone elses control in configuration files we are on the
   wrong train in general. Its bad enouogh to have addresses in DNS
   data, but in config files is even worse. Just imagine a helpdesk at
   an ISP trying to specify the a new v6 address by phone.

   My second objection is that an address is to restricted. A name
   would have among other properties the ability to point to several
   addresses, which I think will be an important feature.

   I understand that you were concerned about priming of the system
   and presumably the present code does it all at startup before
   lookup service is available. But still, I wonder whether it would
   be possible shange this into a lazy evaluation of the forwarder so
   that we could keep it as a name.

2. It will be needed to provide translator service in both directions
   at some point in time. Regardless of how much people are urged to
   keep all DNS data available over v4 transport we will see a mix of
   all three variants (v4, v4+v6 and v6).

   Therefore, obviously, this will get a bit complicated since BIND
   will have to divide its forwarders into two sets depending on the
   direction of translation that is needed and the type of stack that
   is available at the local machine when glue of the wrong type is
   encountered.

3. A question: what happens if there is both A and A6 glue for
   ns.foo.bar.org and they expire from my cache at different times? 

   As far as I understand there is no way of knowing that something
   may have been lost, so I will simply assume that I can only reach
   ns.foo.bar.org over f.i. v4 transport although in fact v6 is
   available. Hence I will (being v6-only) go through a translator,
   even though it really isn't necessary.

I think we will have to live with the translator model for a *very*
long time and therefore we should look at it more carefully than would
be needed for a quick hack.

Johan