Re: host transfers

[Eric Brunner-Williams in Portland Maine <brunner@nic-naa.net>]

>At 5:41 AM -0400 8/8/01, Hollenbeck, Scott wrote (quoting me):
>>  >Does EPP need a registry-to-registry communication mechanism?
>>>If not, does a registry-to-registry communication mechanism need to
>>>exist separate from EPP?
>>
>>As things stand right now, there's nothing that prevents a server from
>>switching into client mode to communicate with another server.

True, but to communicate with a registrar would require that the registrar
also cross dress.

>Indeed.  Eric's suggestion that we have peer-to-peer communications 
>mechanisms allows for (in my mind at least) an even more 
>straightforward relationship between registries using EPP, although I 
>tend to think the difference is a pretty minor one.

Yup, no role-reversal ballet, and entities-other-than-registrars could
include escrow agents, as well as registries of the same, or different
types, e.g., xLDop -> yLTop, where x > y, and y >= g (aka "0").

>Of course, it's not as simple as just saying 'let registries talk 
>using EPP'.

To be sure.

>         ... As currently designed, EPP doesn't provide a mechanism 
>to authenticate against an object without initiating a transfer. 
>Similarly, although the <info> command does provide the needed 
>authentication information, registries may not find it desirable to 
>make all of the data contained within the <info> response available 
>to every other registry.  Similarly, we've discussed in the past the 
>need for registries to communicate changes in the names of registered 
>name servers to other registries.  EPP doesn't provide a mechanism 
>even vaguely like this at present.

Good points.

>I don't think that a more fundamental question has been addressed in 
>the discussions thus far:  is EPP the right way for registry-registry 
>communications?  I haven't thought about it enough to come to a firm 
>conclusion, but my current opinion is that the answer is 'no'.  EPP 
>is, and should continue to be, a provisioning protocol, not a 
>nameserver update announcement protocol or contact object 
>authentication protocol.  Registries need these later types of 
>functions from other registries, and I think that they look 
>differently enough that it will be difficult to 'fix' EPP to the 
>extent that it performs all of these functions well.

Lets answer the "why no <push>" question, preferably in the affirmative,
then consider the scope of to whom registries initiate the transfer of
state to, and which state is transferred.

Oh, and lets make sure we're as generic as we can sensibly be too, as
the addr reg people haven't weighed in previously.

>Jordyn