More comments on "Revisiting Vertical Separation of Registries and Registrars"

[Eric Brunner-Williams <ebw@xxxxxxxxxxxxxxxxxxxx>]

My previous note covered five critical comments about the CRAI-1 model. 
Here I try to be more positive about the model, or differently negative.
1. We have the $75,000 per year number for the recurring ICANN fee, in 
lieu of the $0.20 per domain fee for most, but not all, of the gTLD 
contracts. This number looks wicked high if we think of the CRAI-1 model 
as a defensive buy, or a vanity buy, and absurd, but tolerable if a 
brand management buy, and by the time we get to a large trade union and 
its membership, we get down towards the $0.20 price point.
But is that as big as it gets? Suppose Google "inverts" its search 
engine, and under .google, a CRAI-1 entity, dumps gazillions of search 
strings into its name space, and not just today's search strings, but 
yesterday's, and the day before's, and so on. At this point the 
per-leafnode cost for Google's tree is wicked close to zero, and 
everyone's eyes are glazed. But is the value of all of those leafnodes zero?
At present, the PPC model works by populating jay-random tasted, or 
otherwise trafficked nodes, and here's the key point, pre-existing, 
mostly static content or templated garbage nodes, with tasteful Google ads.
Why should Google searches result in anything other than links to 
.google nodes with all content created on-the-fly?
So I think there are ways to make a great deal of money gaming the 
single-registrant-with-flat-fee model and we shouldn't assume the CRAI-1 
model is just corporations giving email to employees or managing product 
CRM.
2. We have the claim in the CRAI paper that the registrar cost is an 
inefficiency, which I read as the authors of the CRAI-1 model can't 
think of ways to use the DNS to make product with a per-node/per-year 
value greater than the margin charged by the large, highly automated 
registrars. That's point #3 of my prior, unreservedly critical note.
I suspect that there are applications that require a clean namespace 
that have a greater per-node ROI than $1/yr, and since these are clean 
namespaces, PPC is obviously not the application. Actually I'm sure of 
it, so much so that I don't see the registrar margin as a fatal defect 
to the application, and the registrars do market the application, not 
for my value, or the registrant's value, but for the value of their 
dollar margin.
3. While its simply insane to imagine, as the CRAI-1 authors did, that 
an entity could usefully attempt "information security" by not allowing 
3rd party disclosure (and the notion that Corp X, mortal competitor to 
Corp Y, would become ICANN accredited, and pass Corp Y's OT&E, for the 
purpose of garnering critical information through node add/mod/delete 
commands, is the most creative exercise in ICANN fiction ever), it is 
easier to introduce DNSSEC without gTLD registrars for whom the Y tld 
and all its private hoopla, key management inparticular, is a 
distraction from the DNSSEC-free-forever COM honey pot.
There actually are rationals other than registrar disinterest to want to 
be able to impose some coherent policy recursively. Suppose the 
contrary. Suppose the registry has no policy control except over second 
level domains, but not recursively. The only "safe" outcome is a flat 
namespace full of hyphens and driven by search engine key words rather 
than recursive descent on standard label taxonomies. That way lies 
HOSTTABLES, and we engineered our way out of that a long time ago.
4. "Corporate TLDs" will be a bonanza for consultants, and while they 
will have an adverse affect on the IANA root's ability to accommodate
future for-profit and non-profit public services, point #1 of my prior, 
unreservedly critical note, the current recession in the dollar and euro 
zones will be at least as destabilizing on the market as the bubble bust 
at the beginning of the decade.
A problem the CRAI-1 author overlooked is what is the correct metric to 
vet a would-be-registry-operator for the typical CRAI-1 class of 
on-going operations? At Rome I pointed out to that Board that we who 
drafted 2000 round applications, and .net and .org redel bids, were 
forced to lie about the actual capitalization and the actual 
capabilities of our organizations in our bids, and less fiction would 
(a) make application drafting less expensive, (b) make application 
evaluation less expensive, and (c) increase the honesty of the system, 
each of which would probably be a very good thing.
So does a defensive buy have the same registry-operator requirements as 
a brand buy? How about a provision-the-products-and-employees buy? How 
about a vertical application with a sparce, but highly secure tree? How 
about a wicked bushy tree with no added bells and whistles, other than 
the minor nuisance of being able to on-the-fly create custom content at 
bazillions of leaf nodes? It seems to me the real requirements range 
from the original .museum kitchen table PC to VGRS's Atlas platform.
The  CRAI-1 model is a mix of good and bad, sold as unleavened good, as 
consultants usually do. A bazillion soap and cereal labels stuffed into 
the IANA root would leave us all looking at the Google keyword API like 
it was the next best thing, and our discomfort would not cost a moment's 
sleep for the scores of people who would have made a killing "helping" 
brand managers move into the best ad buy ever.
Eric Brunner-Williams
Disclaimer: I work for CORE Internet Concil of Registrars as CTO, and I 
have to plan for both possible outcomes, so read my comments with those 
interests in mind.