Re: [dnsext] the same in old days, was making names the same NEED protocol changes?

[Ted Hardie <[e-mail]>]

On Mon, Feb 28, 2011 at 7:07 AM, Phillip Hallam-Baker <[e-mail]> wrote:
<snip>

> So attempting to make two names resolve alike without change at either
> the application client or the application server is a fools errand.

I'm not sure how you see this split based on the text above.  My
personal take is that if you want two names be "treated the same" by
an application, you can do one of three things:

1) Make one name refer to the other, transforming the problem of
"treating the same" into "pointing to the canonical name".

2) Provide for all the names equally, so that the same resources are
available at each name (this means not just things like
name-to-address mappings but also certificates with the names used on
them etc.).  This is expensive and difficult, but there may be some
methods to reduce the pain (subjectAltName, provisioning-based
mirroring of resources, new protocol methods).  I think a lot of the
discussion so far has been of the "are there useful DNS mechanisms
that could reduce this pain"?

3) Create a method that allows applications to know which names should
be treated as equivalent.  The difficulty here is primarily that any
names which are not within the same administrative boundary cannot be
treated as equivalent without the records related to both/N
administrative boundaries confirming the equivalence. So the
application would have to understand how to determine the
administrative boundaries, check for the confirmation of equivalence,
and then correctly apply whatever behaviors result from the two names
being the same.  This is far from trivial, as the
applications/protocols may have no defined behavior for "treat these
as the same".  Knowing they are to be treated as equivalent doesn't
help a lot for applications with no sense of what behavior should
result.


> is not going to work because that is simply not how the applications
> see the world.
>
><snip>
>
> If we decide we can change the application client, this whole problem
> becomes pretty straightforward. Either adopt the 'did you mean'
> pointer style approach or allow domains to nominate mappings of one
> charset to another.
>
Please do not use "mappings of one charset to another" to represent
this problem.   At this point, we are all mostly talking about the
Unicode character set and how to map labels derived from sequences of
Unicode characters.

regards,

Ted Hardie