New ARIN ipv6 allocation policies

[Iljitsch van Beijnum]

On 3-sep-2006, at 5:46, David Barak wrote:

>> As we've learned from IPv4, I fear swamps are
>> forever...

> But so what if they are?  A single /32 of /48s is an
> awful lot of space

Yes, but that's irrelevant. Fortunately, even though it looks idr  
isn't going to be any better with IPv6, it still solves the address  
shortage problem and then some.

The problem with a swamp is that it forces people to carry more  
routes in their routers than necessary, irrespective the amount of  
address space in the swamp.

> - Assume one PI block per
> organization (because if an enterprise can actually
> run out of a room in a /48, they're probably really a
> service provider anyway...), and further assume that
> all of the companies who have ASNs today immediately
> request a PI block from the swamp.  At that point,
> you're at what, 35K unique routes?   What equipment
> will have trouble with this?

IPv6 routing needs more memory because the addresses are longer and  
there are now two next hop addresses (global and link-local) so there  
could be surprises here, but I grant you that 35741 IPv6 isn't beyond  
the capabilities of current technology.

But this isn't about today, tomorrow or next year. This month, we're  
celebrating 25 years of IPv4 and we know how hard it is to transition  
to IPv6, so IPv6 will have to last us until at least 2030 and  
probably longer. In IPv4 there is a hidden limit on the number of  
multihomers: you need to qualify for at least a /24. In IPv6,  
everyone can get a /48. So there is a potential for many more  
multihomers than there are in IPv4.

(The current ARIN policy requires someone who wants an IPv6 PI block  
to qualify for an IPv4 block first, which is hard in ARIN country,  
but this isn't very compatible with (for instance) the way the RIPE  
NCC works, so I think it's very likely that now that ARIN allows PI  
the same will happen in other regions with fewer limitations than  
there are now with ARIN, and after that the people in the ARIN region  
will want to get rid of some limitations and we'll see a race to the  
bottom.)

> According to Wilhelm & Uijterwaal, the growth rate of
> actual RIB-appearing ASNs is ~ 200 per month,

So why is the growth in prefixes in the IPv4 table so much higher?  
It's even higher than the growth in the number of prefixes given out  
by the RIRs.

Probably: people who can't be bothered to aggregate, and people who  
want to do traffic engineering by deaggregating. I see no reason why  
those behaviors wouldn't translate to IPv6 to at least some degree.

> and we
> shouldn't expect to run out of 16 bits for the next 7
> years or thereabouts.

Let's see:

+------+----------+
| year | sum(num) |
+------+----------+
| 2000 |     3520 |
| 2001 |     3777 |
| 2002 |     3367 |
| 2003 |     3335 |
| 2004 |     3563 |
| 2005 |     3870 |
| 2006 |     2999 |
+------+----------+

There are currently 35741 AS numbers in use and 1537 reserved so that  
leaves 28770. This year we seem to be heading for 4500 AS numbers  
used so 7 years seems reasonable assuming no further increase in the  
growth of AS numbers used per year, which seems a strange assumption  
to make based on the above.