[ppml] PI addressing in IPv6 advances in ARIN

[Owen DeLong]

--On April 14, 2006 12:20:06 PM +0200 JORDI PALET MARTINEZ
<jordi.palet at consulintel.es> wrote:

[snip]
> However, I want to balance this with the medium-long term implications
> created in the routing table and with the time needed to build and deploy
> a better technical solution (or several) which is accepted by the
> community.
> 
I think we first need to define what we consider a solution... See below.

> So my proposal basically is about having PI now everywhere (once ARIN
> adopt it, is unfair not having it in other regions), but those PI
> allocations for multihoming should be temporary and those address blocks
> returned to the RIRs some time (lets say 3 years) after the new technical
> solution is declared as a valid one.
> 
I would not actually support this idea.  The whole point of having PI
space is to have the addresses for a long-term.  Having a timeframe for
return would simply restore the same barrier to entry that existed
prior to passing the policy.

Other RIRs are free to implement whatever v6 PI policy they feel is
appropriate for their region.  I would support a globally standardized
v6 PI policy along the lines of ARIN 2005-1.

However, I would like to argue that if the new technical solution will
benefit from the return of this address space, it is most likely not
truly a solution, but, instead, another clever hack piled on top of
the existing set of hacks.

I suppose if someone found the magic bullet to make geotopological
addressing really work, that might qualify.  However, I have very low
expectations in that area.

Absent that, any true solution will involve making the size of the routing
table independent of the number of PI (or even PA) blocks issued by
the RIRs or will make the size of the routing table practically
irrelevant.

I know this isn't the easy solution, but, we need to look long and
hard at the way we do things.  I think that solving these problems
is going to require a significant paradigm shift.  Assuming that we
can use IP addresses for both end system identification and for
routing topology indicators is how we created this problem.  I don't
see solving it without breaking that assumption, at least at the
interdomain level.


> At this way, on the long-run, we will not have routing table implications,
> but we allow now the people that want to move ahead only if they have a
> multihoming solution doing so.
> 
If you think there is a possible solution (a real solution, not just
a hack that postpones the inevitable at the expense of usability
like CIDR did), then I'd like to hear what you are thinking.

> This 3-years time for getting a multihoming network back to the new
> technical solution (once adopted) is enough time, I think (it could be
> changed to 5 years if needed, or whatever), so nobody today see the
> temporarily of the proposal as a showstopper to go for it now.
> 
I think you underestimate the momentum and requirements of the modern
enterprise if you believe that to be true.  Any capability available
in v4 that is not available on at least equal or better terms in v6
is a deterrent to v6 deployment.

The ability to get permanent addresses which do not have to be returned
when you switch providers or renumbered on a schedule determined by
some external organization is a major example of such a capability.

Owen


-- 
If it wasn't crypto-signed, it probably didn't come from me.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 186 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20060414/04744b32/attachment.sig>