[address-policy-wg] PI concept in general, was IPv6 PI resource question!

Daniel,

On Tue, 2011-02-15 at 20:41 +0100, Daniel Roesen wrote:
> We're allowing PIv6 for "multihomed" (whatever that means really) sites.
> Those could "just simply" use PA space from any ISP they connect to. Why
> do we make Internet multihoming special compared to Internet-plus-others?
> With the "multiple IPs per device" argument there cannot be PIv6. I
> thought we've left that behind by now.

So I was wondering "why do we allow PI for IPv6 for multihoming?", and
realized "that's how we do things in IPv4". AIUI people don't consider
it reasonable to get one ISP to carry routes from another on behalf of
their customers. Fair enough.

----

In this IPv6 PI discussion people say it is too expensive to be an LIR.
This strikes me as weird... even cheap network gear costs way more than
the LIR fee, and certainly the personnel costs are many times that, even
in countries where technical folks are poorly paid. But anyway....

On the other side we hear the concerns about unchecked routing table
growth. This has been addressed many times; it is usually presented as a
basic "tragedy of the commons", where the costs of the increased size
are borne by everyone but the benefits go to each network which
advertises any given route. 

This position is directly refuted by Geoff Huston who claims that memory
is cheap, what we really need to care about is update frequency, which
is not actually closely related to the total number of routes today but
is more related to a number of problematic AS (see chart at end):

http://bgpupdates.potaroo.net/instability/bgpupd.html

Further, in IPv6 we're much less likely to have multiple routes per
organization (although of course as companies merge and diversify this
won't be 100% true). Plus we have the advantage that as IPv4 gets more
scarce, it is likely that the number of addresses per advertisement will
get smaller, so the IPv6 will be able to use whatever technological
means is used to keep IPv4 creaking along.

So... it may be there is no routing reason to hold back on PI blocks.

----

Is there another reason to want to limit PI assignments? There may be.
Certainly the folks over in abuse-land think we should have super strict
controls over all addresses, and revoke them at the first hint of
naughtiness. Or there may be administrative reasons - an entirely
different kind of infrastructure is needed to manage hundreds of
thousands of low-fee (well, NO fee now) PI-holders than thousands of
high-fee LIRs. I don't know.

So... do you think we should simply remove all restrictions for IPv6 PI
and issue space if someone wants it, period? If not, what possible
restriction would there be?

----

Finally, I think what bothers me about PI in general though is that once
the initial assignment is made... that's it. At least with RIPE 452 we
have *some* contractual relationship with the PI holder, but that's
about it.

I guess the "LIR middleman" model comes from the mists of time, long ago
in history. It certainly reduces the burden on the RIPE NCC, but it
never really struck me as sensible. After all, isn't the whole point of
PI space that you don't want to have to depend on an LIR?

--
Shane