[ppml] [address-policy-wg] Those pesky ULAs again

[michael.dillon at bt.com]

> Conversely, it would be rather trivial to add random address 
> assignment to IPv4.  ICMP router solicitation, auto configure 
> into the subnet, randomly assign a IP, look for a collision.  
> AppleTalk did it 15 years ago.  Worked good for small 
> deployments.  Would be fun to write the code and check it 
> into several distros....

If you don't, then somebody else will.
Remember when I said that one strategy to mitigate IPv4 exhaustion is to
share addresses between regions, i.e. an American small business could
use the same allocation as a Korean small business and a French small
business. It is unlikely that they would ever need to communicate much
outside their regions. And the RIRs could step in an manage such
allocations to prevent too many people using the same address and to
ensure that the users understand the limitations of such shared
addresses.

However, if your devices could pick a random IPv4 address, check whether
or not it is Internet reachable, then check routability to some major
sites, and then decide to use it...

Note that this would also solve the RFC 1918 exhaustion problem that
many large enterprises are struggling with.

In general, I'm in favor of making all possible mitigation tactics
available and letting the market decide. Your random allocation and test
tactic. The class E address tactic. The expansion of RFC1918 address
space tactic. And others yet to be though of.

--Michael Dillon