[ppml] Proposed Policy: IPv6 Direct PI Assignments for End Sites

[Craig Huegen (chuegen)]

David Williamson wrote:

> owen beat me to it.  I agree, and on the same grounds.  Personally, I
> think this problem has become somewhat intractable.  Parts of the
> community really want to see IPv6 become a reality in the ARIN region,
> which will require some sort of PI policy.  Others are concerned about
> the impact on routing slots and other resources (or, more cynically,
> their organization's bottom line).  Until the various

I'm concerned about both, to be honest, and a balancing act is required
here.

Since once off the ground, IPv6 appears to be headed for a fairly long
life, we do have to be concerned about the scalability of the
infrastructure.  In addition, beyond the basic DFZ routing table memory
that I hear concerns about, I share concerns about scalability for other
services (e.g. security) inside the network (mechanisms such as ACL's
via TCAM, etc.) and the technology needed to drive that.

We won't need to worry about any of that if we never get adoption.  In
my opinion, without enterprise businesses adopting IPv6, it will likely
remain a research and hobbyist playground.  No evidence suggests this is
solely an ARIN problem, either.

I would like to re-emphasize the notion raised previously that a
requirement to use PA space is a de facto perpetual contract with an SP,
for an enterprise that has a significant number of subnets.  A small
business with 6 or 8 segments isn't impacted in the same way as a large
enterprise with 60,000+ subnets is impacted.  For those large
enterprises, it is simply unacceptable to use PA space.

I have not observed a strong feeling in these policy discussions against
the idea of PI space; most of the arguments I see are related to where
we place the line in the sand for eligibility.  I offer that the line
needs to be drawn based on the pain of renumbering PA space vs. the
impact on the DFZ infrastructure.  In a relatively large corporate
network that I have an association with, renumbering tens of thousands
of subnets is extremely painful and it is unlikely that we would
extensively deploy IPv6 in the near-term knowing we would have to
renumber every one of those subnets in case of a service provider
change.

When comparing this versus existing IPv4 models, I don't feel that a /22
or /21 generates sufficient renumbering pain to warrant assignment of PI
space.  I sit on the fence when it comes to /20, and I rather like the
idea of using /19 as the guideline (personally).

> I generally support the current version of 2005-1.  As someone once
> noted to me, a good compromise is the one that makes all parties
> slightly angry.  That proposal seems to have that property...hence my
> support. 

With all that said, I think the most important point here is that
adoption of IPv6 in the near term will require some type of PI space.
I'm inclined to support / vote for any proposal that implements PI space
in the near term with reasonable parameters.  I want a longer term
solution -- whether it be id/loc split or some other method -- that
would eventually phase out PI space.

/cah

---
Craig A. Huegen, IT Solutions Architect       C i s c o  S y s t e m s
IT - Intelligent Network Solutions                  ||        ||
Cisco Systems, Inc., 400 East Tasman Drive          ||        ||
San Jose, CA  95134, (408) 526-8104                ||||      ||||
email: chuegen at cisco.com       CCIE #2100      ..:||||||:..:||||||:..