[ppml] Policy Proposal: IPv6 Assignment Guidelines

[David Conrad]

Steve,

On Aug 17, 2007, at 2:24 PM, Steve Bertrand wrote:
>> There is no shortage of IPv6 addresses and if the world
>> ever does run out of IPv6 addresses, we will all be dead when it
>> happens.

Michael is right about one thing: we'll be dead because the Internet  
users of the world will unite and rightfully hunt us all down and  
kill us for being so inexcusably stupid as to burn through 2^128  
addresses.  :-)

> What happens if someone implements something in the stack that would
> allow a device on a network to iterate through a billion addresses
> randomly every few minutes

I think you're not really grasping the magnitude of 128 bits of  
address space.

If you cycle through 1,000,000,000 addresses every "few" minutes  
(where few = 3), it will take over 105,000+ years to exhaust a  
single /64.

> What if every single IP enabled device on the planet did this?

Assume 7 billion people.  Each person could have 2,635,249,153 IP  
enabled devices, cycling every 105,000+ years.

The point of this meaningless exercise is that it is pretty hard to  
come up with a rational way to burn through 128 bits of address space.

However, people are finding interesting ways to be irrational.

Discussions about the length of assignments are largely pointless.   
Even if the RIRs all went back to the IETF addressing  
"architecture" (to use the word _very_ loosely) and allocated /48s to  
enterprises, there are a total of 35 trillion /48s in the IPv6  
address space in the 1/8th of the address space the IETF has  
designated for unicast.  This works out to over 5000 "enterprises"  
per person (again, assuming 7 billion people).

The real problem isn't IPv6 address exhaustion, it is the risk of  
overwhelming the routing system.  I would worry that allocating  
longer prefixes under the theory that it is less wasteful of address  
space might increase the chances that those longer prefixes would  
show up in the routing system.

Regards,
-drc