[Ietf-not43] extensibility

[Eric A. Hall]

on 7/25/2003 7:01 PM Edward Lewis wrote:

> It's not a matter of what's best for the user vs. what's best for the 
> provider.  Between FIRS and IRIS, well, let's assume they both 
> provide the same service.  In that case I'd say it is valid for the 
> providers to lean to the one that offers it more cheaply.  Because 
> the cheaper one is chosen, in the long run every one wins.

I disagree with both the premise and the conclusion.

Given that user-to-user applications can be expected to be deployed that
reuse the chosen transport, namespace, messaging, etc., the unavoidable
conclusion is that what's best for the user *IS* going to be the most
important. What is easy for them to deploy and manage, what offers the
most integration into other services, these are the criteria we should be
focusing on, since those will ultimately determine whether this is
actually a successful broadly-deployed service or if its just a more
expensive version of whois that is once again limited to being offered by
a handful of delegation bodies.

Furthermore, as far as the expense argument goes, I think that a critical
analysis on that point would actually favor FIRS. It may be cheaper for
delegation operators to reuse the provreg stuff (although I'm not even
convinced of that; just that it's more convenient), but for the average
user the initial cost won't have already been underwritten. I mean, LDAP
gives them a pre-networked database that is tightly integrated with a
transport service already, provides integrated authentication and
authorization services, has broad support from the vendor and support
communities, etc. I can't even find a BEEP module on CPAN, and remember
we're expecting the community to stitch together unspecified databases and
XML together and security and all of that everytime they want a new
application in the service space. Cheaper? LDAP definitely.

Then we get into the real meaty stuff, like the number of RFCs that have
gone into LDAP, and how long do you expect it would take to replicate all
of that in databaseX-beep-xml. Or, which do you suppose the PKIX community
would prefer, starting over with databaseX-beep-xml or LDAP. And numerous
other similar arguments.

Really what the favoritism towards IRIS boils down to is a preference for
convenience. Now we've got a hammer, let's go find some nails. That might
even be okay -- I'm certainly favoring LDAP because of convenience reasons
too -- except that provreg is a closed-community service. By that I mean,
the providers are the consumers and vice-versa, and supporting millions of
application agents probably wasn't much of a design consideration, while
that really has to be this WG's focus entirely (even if no user-to-user
applications are ever developed, we'll be looking at that scale from
programmatic lookups alone anyway).

> This is why I raised the question about the name space and 
> extensibility.  Is it just that I don't see the advantage of 
> domainComponents?

I don't see what domainComponent has to do with any of this. So the
linkage to a universal, managed namespace is transparent, so what?

-- 
Eric A. Hall                                        http://www.ehsco.com/
Internet Core Protocols          http://www.oreilly.com/catalog/coreprot/