US CODE: Title 15, Chapter 1, Section 2.

[Howard C. Berkowitz]

Karl,

This is a very useful clarification.  It brings up several ideas/questions,
not necessarily ones that CAN be answered within the scope of this list,
but nonetheless valid areas to consider.

At 2:18 AM -0800 2/1/97, Karl Auerbach wrote:

>I think I figured out why there is so much steam being generated.
>
>My notion of "routability" doesn't mean that at the instant ARIN says
>"here's a number" that one can go out, apply power, and magically expect
>packets from the outside to find their way to the new block.  That's not
>possible until the block has topological context.  I suspect that folks
>were thinking that I was proposing some magical, impossible thing or that
>the ARIN folks were going to have to configure the assignee's bgp or
>something like that.

It was coming across that way.

>
>My use of the term "routability" was ment in the prospective sense -- that
>once a block was actually given topological significance -- i.e. that it's
>exchange points with the rest of the world were determined -- then there
>would be no artificial limits on the acceptance of that new block.  (By
>artificial I mean things like "ISP X won't accept your advertisements
>because your block is too small.")

Let me make a distinction intially between ISPs and users, and, for
discussion only, make the assumption that an "ISP" provides transit service.

It does appeal to me, as a first thought, that if an ISP meets the RFC2050
or successor qualifications for allocation, that there is a presupposition
of routability.

Things that complicate this, however, fall into the transit vs. peering
problem.  Let's say a new ISP in a major metropolitan area applies to ARIN
for a good-sized block -- let's say a /17 -- that it has real business to
justify.  I think we would all agree this is a large enough block of
customers such that it serves the global Internet well to have them
globally reachable, and vice versa.

But remember that I said this is a metropolitan area provider.  If they
connect only to their metropolitan exchange, do national providers have an
obligation to announce the /17 globally?  If so, how does one deal with the
potential economic impact that this provider is getting
national/international transit services for which it does not pay?

>
>In other words, membership in ARIN, or perhaps even the use of an address
>by an ISP of a block carved from a larger ARIN allocated block might need
>to carry with it an obligation on the part of that ISP to honor all other
>ARIN derived allocations.

Subject to the economics of transit, my first reaction is that this is a
not completely unreasonable assumption.  But the economics of transit may
make it infeasible.

Now, let's turn to a different case.  An end user organization -- let's
make it a medical lab providing life-and-death information -- has less than
200 hosts, so it can only justify a /24.  Due to its mission, it is
extremely concerned with being reachable, and wants to multihome to three
or more providers. Let's assume this company has fully clueful routing
people.

In other words, I have set up what I would consider the ideal justification
for a small organization to get provider independent space that needs
global advertising.  Yet, if I give this organization a PI /24, I set a
precedent for assigning such blocks.  There is much more demand for /24,
and quickly all the routing table growth (including flapping) issues arise
again.  Many small organizations don't have the knowledge that does the
ideal firm, and want PI space or multihoming simply because they heard it
was a good idea.

Should ARIN grant PI space selectively, perhaps subject to criteria such as:
    1) There are specific business needs for multihoming
    2) The organization commits to implement multihoming within xxx days
       of being assigned PI space
    3) The organization has staff, or contracted consultants, who have
       demonstrated BGP expertise
    4) The organization commits to renumbering-friendly network design, so
       if in the future the PI requirement disappears, there is no renumbering
       pain disincentive to releasing PI space

Again, I don't have answers, and suspect that many of these questions are
outside scope -- they are more appropriate for PAGAN or other lists dealing
with fundamental allocation policy.

Your clarification helped.  Thanks.
>
>By-the-way, I wasn't proposing a "free for all", only suggesting that the
>possibility exists for the net to devolve into clouds of competing network
>numbers.  And I did say that I considered that to be a dangerous future.
>

I think this would be dampened very quickly by any providers wanting to
survive.

Howard