Re: [dbound] RDBD Questions

[Stephen Farrell <[e-mail]>]

Hiya,

On 23/07/2019 06:37, Andrew Sullivan wrote:
> Hi,
> 
> Same disclaimer as before.  (Sorry that it gets dull, but we really
> want to make sure the ISOC/IETF LLC relationship is a bright line
> before we stop.)

That's ok. A #include form might work for follow-ups:-)

> 
> On Tue, Jul 23, 2019 at 05:52:17AM +0100, Stephen Farrell wrote:
>>
>> On 22/07/2019 23:57, Andrew Sullivan wrote:
> 
>>> The problem the draft is trying to address, however, is _not_ a DNS
>>> relationhip.  Instead, it's an attempt to affirm that there is a
>>> _policy_ relationship between two domains.  This is reasonably clear
>>> if you look at the use cases in the bullet list.
>>
>> Well, we're trying to be agnostic as to the semantics of
>> what the relationship is. The logic being that if we get
>> the mechanism right, additional tags can be allocated later
>> when specific semantics are in play. So I'm not sure if
>> describing that as a policy relationship is quite right.
> 
> This is possibly the part that's balling me up, and possibly the place
> that I need a better understanding.  I think that inside a protocol P,
> there are two things you can state:
> 
> 	1.  This is how A & B relate inside P; or
> 
> 	2.  This is how A & B, which are objects of P, relate outside P.
> 
> I claim that any operator of P has exactly those two choices, and that
> from the point of view of the operator of P everything in (2) amounts
> to policy.

I agree rdbd matches case 2.

I think calling that "policy" is maybe not a good plan though
as people seem to generally infer that policy issues are
always extremely weighty matters that require lots and lots of
complex mechanics.

> For the particular thing that RDBD is doing, the draft approaches this
> from the point of view of "things I want to put in the DNS about an
> obect of management for the DNS".  From the DNS's point of view,
> that's a _policy_ expression, whatever you might think it applies to.
> You can be agnostic all day long about semantics, but you are not
> changing the semantics of the DNS and therefore inside the protocol
> you hope to address, you're expressing policy.
> 
> I think this is a very fine thing to do, please note.  I just think
> that, given the way DNS works, it has to be symmetrical or you won't
> be able to express what you think you can.

Hmm. Two things I guess: 1) I fully admit my relative ignorance
of DNS, so you may be right, and 2) I still think you're maybe
wrong:-)

If we're dealing with an rdbd tag of 1 (some, unstated, positive
relationship), then I think the rdbd mechanism can still be
uni-directional even if the real-world relationship is not. And
using the mechanism twice, once for each direction, is entirely
possible. So a uni-directional mechanism seems to me to be a
better stake in the ground.

Second, if we're dealing with some other putative tag value, say
with some semantic related to ownership of the real-world entities,
then that's an inherently asymmetric relationship, which again
leads me to think that a uni-directional mechanism might be best.

All that said, you're right that this is a topic to mull over,
and (for this to work out well) we need to address it in some
way that doesn't create big barriers to deployment.

Cheers,
S.



> 
> Best regards,
> 
> A
>