Re: requirements draft revision

[Joe Abley <jabley-ietf@automagic.org>]

On Fri, Jun 29, 2001 at 01:36:31PM -0700, Tony Hain wrote:
> The reason I have been hammering on this is there are some underlying
> assumptions everyone has about what these terms mean, and buried in those
> assumptions are mechanisms for what needs to happen to make things work.

I agree, if we're going to have definitions of the terms we use
in the requirements doc, we need to have them defined clearly. I have
been trying to make the definitions as solution-neutral as possible.

> The reason for the NAT question was trying to distinguish the difference
> between an individual dialing in vs. a larger organization. Basically, what
> does F managing address space have to do with the discussion of C being a
> transit or not?

The idea was to re-use an established term (that definition comes
from RFC1918) as a substitute for "autonomous system" which is the
term used in the previous draft, and which implies routing and BGP.
The reference to address space makes more sense in an RFC1918 context
than in this one, perhaps, but the kind of entity I wanted to use
the word "enterprise" to refer to is the same, I think.

> Does any collection of routed subnets constitute an
> 'Enterprise'?

No. "Any collection of routed subnets" could mean anything, from a
crossover cable to the entire Internet. The definition of "enterprise"
used in the draft is much narrower.

> If so do we care about all of them, or only the ones that are
> trying to punch a policy through C? Is there a way to distinguish those
> cases?

The word "enterprise" is only really being used to describe an
entity which is able to multi-home. We need to call those entities
something. That's not an adquate description, however, since we
need "enterprise" in order to defined "multi-home".

"transit provider" is a role that an enterprise may play in relation
to another enterprise: that of providing connectivity to the Internet.

> The fundamental point I was getting to with that last set of questions was
> trying to really nail the definition of 'transit'. The last line of your
> response implies single directionality. Why is C not a transit provider of
> B? (hint: the term 'customer' may not appear in the answer)

Heh :) I take your point. I guess C *is* a transit provider of B by the
definition in the draft. How about if we s/to the Internet/to the global
Internet/ as Randy suggested?

> If the Internet were a single rooted tree the single directionality
> definition would work, but the Internet is a mesh, so the definition is a
> mess. Are you really trying to identify a term for the set of transit
> networks where one of the attached parties is a stub?

The Internet is, in general a mesh, but in practice it's rarely difficult
to identify who is "upstream" and who is "downstream". If we insert the
word "global" as above, I think the definition will be very familiar
to most operators. To drill the concept home, we could always add some
diagrams.

On the other hand, if you have a better definition, let's hear it.


Joe