[ppml] Policy Proposal 2005-1: Provider-independent IPv6

[Christopher Morrow]

> > It appears that most people at the last want to make a policy for policy's
> > sake.
>
> I disagree.  I think most people are in favor of 2005-1 because they
> recognize that this PI policy is better than no PI policy.
>

I think, actually, it was more: "we need to help jumpstart v6, nothing
would do that better than making folks that need to deploy it not be
tied to a single provider or broken multihoming schemes"... but that's
probably not horribly relevant anymore.

> > There are many other things that haven't been looked at in addition
> > to just IPv4 and IPv6 tables. I haven't heard or seen any studies of the
> > impact that VPNs have in addition and even get more nervous when
> > discussions on PEv6 are brought up.
>
> I'm not sure what special impact PI space would have on VPNs or PEv6.  Can
> you elaborate?  In particular, do you have any idea how many PI routes it
> would take to start breaking either technology (given current hardware)?
>

So, I think the problem might be: Someone has a large IP network, they
overlayed a largeish 'VPN service' on that network. What routing
growth problem are they having? or will they have with ipv6?

If you use some form of the numbers from tony/jason/geoff and try to
predict current v4 table + 'current' v6 table growth what impact will
this have on the above example? I'd say that you might be able to stay
the VPN routes mean a 'static' increase on whatever the math above
shows you as a growth rate.  At the very least, the additional growth
rate from the VPN routes is going to be highly provider dependent. (so
hard to predict globally)

> > As others have mentioned, historically temporary solutions aren't. I
> > believe that we are making the same mistakes as we did when v4 was first
> > rolled out with a /48 out of a reserved /44 per PI request. How large is
> > this PI swamp that is being proposed?
>
> The number of PI allocations will be small, at least at first.  The number
> of entities that have gotten IPv4 PI is small (less than 10,000 IIRC), and
> you have to qualify for IPv4 PI to get IPv6 PI under 2005-1.  Hardware
> will continue increasing in capability, so as long as we keep an eye out
> for problems we can make changes to policy before we start causing issues
> for operators.
>

I think I agree with Aaron here, they just won't ever come back...
there are legacy /24's assigned that no one can recover today. there
will be 'legacy' /48's (or 44's or whatever the decision on size is)
assigned tomorrow and never recovered.

> > If 2005-1 is repealed, how will the space be returned without litigation?
> > Odds are that it won't and we'll be forced with it with no recourse.
>
> And what if the space isn't returned?  Say 5 years down the road a new way
> of multihoming takes off, and we completely repeal 2005-1.  All the
> existing PI holders keep their space, though, and space is only returned
> through attrition.  The number of PI routes in the routing table stops
> growing, but the capacity of routers keeps growing as it has since day 1.
> Before too long, router capacity so far exceeds the capacity needed to
> carry the PI routes that everyone forgets there was ever a problem.
>

that is a nice utopia, I think it's several generations of equipment
off in the future though :(