[arin-ppml] Policy Proposal: Dedicated IPv4 block to facilitate IPv6 deployment

[Paul Vixie]

> ... constantly moving target and getting any accurate measure
> of any of those three numbers would be a challenge ...
> 
> ... My personal suspicion: ...
> 
> ... Probably more than ...
> 
> ... I also suspect ...
> 
> ... My rough guess is ... However, I believe that ...
> 
> ... Further, I believe that ...

so, back to the topic at hand, is there an experiment design in the offing?

guesses, suspicions, level of challenge, probablies, and beliefs won't help
us.  it doesn't matter what we suspect, guess, or believe.  what matters is
what we can measure and what we can prove.

is there an experiment design in the offing?  we all know what one another's
suspicions, predilictions, guesses, fears, hopes, and dreams are.  what we
need to know how is, what can we measure and how, what can we prove and how.

straw man: grab a dozen unrouted /28 blocks from ARIN's inventory having wide
range of first-octet, pick a dozen existing (routed; known to be old and
stable) places; arrange the machinery and software to be able to do ping tests
of some known reachable addresses within every prefix in today's global
routing table; (ping means icmp-echo but could also include a TCP SYN to port
65535, and in all cases we need to remember the address of the gateway who
sent us an icmp response); variations should include advertising each /28
and re-running the ping tests after 10 minutes, 100 minutes, 1000 minutes,
and perhaps faking a route flap to see how that changes the results; compare
the ping test quality of these /28's against the ping test quality from the
existing/routed/old/stable prefixes run at the same time, to filter out the
remote-end reachability problems and hopefully leave us with /28-nonpropagate
events.

ping tests of this kind are incomplete.  we also need to institute ping and
TCP/80 from within enduser networks, and it will have to be more than a dozen.
but because of the difficulty of designing and executing that experiment, i'd
say we should ask for the ping test results first, since if they are dismal,
then marty's proposed amendment will fail and we won't need to know what the
inbound reachability of these prefixes would have been.

who can improve on this?