Re: [v6ops] PI [ULA draft revision #2 Regarding isolated networks]

[Tore Anderson <[e-mail]>]

* Brian E Carpenter

> Tore,
> 
>> We have a few customers in the same situation. So we obtained a PI
>> prefix for them, 
> 
> The important words there are "a few". As long as the numbers are
> reasonable, this scales. When the numbers cease to be reasonable,
> it doesn't scale. The estimate I made some years ago, based on
> a little research into statistics in a few countries, was that
> there must be about 10 million small or medium enterprises in the
> world. We don't know how to route 10M prefixes in BGP-4. So
> somewhere between "PI for a few customers" and "PI for every
> enterprise", we have to stop.

For sure, it is a very small minority of my customers who have this
requirement. The vast majority are perfectly happy to be assigned
prefixes out of my PA block. This goes for IPv4 too, BTW.

I seriously doubt that a significant fraction of those 10M enterprises
asking for a PI prefix (or multihoming at all) is a realistic scenario.
So I'm not too worried about IPv6 PI to be honest, we survived IPv4 PI
so far and I think we'll survive IPv6 PI also, even if adding a few
"new" PI holders who in IPv4 multihomed using PA assignments from their
upstreams + NAT44 + RFC1918. Especially if homenet comes up with an even
more attractive solution than PI for that specific use case soon.

Tore