We want the absence of (qname, qtype) to be cached by resolvers who follow
a delegation to an omniscient as112 server for all (qname, qtype).

Given that requirement the thought was that NOERROR/no data and NXDOMAIN
were equally plausible. However, I see now that the lack of "no children"
signalling has the potential to cause increased query load, since resolvers
will re-query for children despite an earlier query for a parent. My
feeling is that this is not a big deal and the ability to add/drop with no
coordination with server operators represents a greater win, but I have no
science behind my words.

The current scheme (witness Warren's code) seems to do this (modulo extra
queries for children) for all (qname, qtype) where qtype != SOA. Which is
not perfect, but, I think pretty good. We could servfail on qtype == SOA I
guess, which is crazy nasty but which I think for all practical purposes
would work.

I like your other message's "minimal response" notion, though. When I'm not
sitting on a plane headed for a beach, I will play with that.


Joe

Aue Te Ariki! He toki ki roto taku mahuna!

On 2013-02-22, at 16:53, P Vixie <[email protected]> wrote:

Sorry to be late on this, missed it earlier.

Nxd says there is no such name, no matter what the type was, and there are
no children. No data/noerror says there are either other types or children
or both. We know what the truth must be and we know which implications we
want the requestor to follow. Right? Is there any doubt at all or any
ambiguity in what we want said? ... Paul

Joe Abley <[email protected]> wrote:
>
>
> On 2013-02-22, at 07:57, Paul Vixie <[email protected]> wrote:
>
> if we can't return nxdomain, then i'm opposed to the omniscient spec,
>> and we should continue as before, enumerating on the responding servers
>> every zone to which we wish to delegate.
>>
>> noerror/nodata is the wrong answer.
>
>
> It does smell a bit wrong, but I'm interested in more details of why you 
> think that if it's more than just the smell.
>
> I'll note that the proposal is to assign new nameservers to be omniscient, 
> and not to convert the existing ones. We are then in a position to delegate 
> other local-zoneish zones to the new servers and review the impact.
>
> So, the suggestion is not to replace the existing AS112 servers (blackhole-1,
> blackhole-2 and prisoner) out of the gate. Any such replacement would happen 
> later, once there was some degree of comfort that it was safe to do that.
>
>
> Joe
> ------------------------------
>
> DNSOP mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/dnsop
>
>
-- 
Sent from my Android phone with K-9 Mail. Please excuse my brevity.
_______________________________________________
DNSOP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dnsop

Reply via email to