[ppml] Policy Proposal 2007-6: IPv4 PI minimum size change

[David Williamson]

On Tue, Mar 06, 2007 at 11:47:17AM +0900, Edward Lewis wrote:
> >Rationale:
> 
> >While experiments indicate that a few ISPs still try to filter at the
> >/22 boundary, I have been repeatedly told that most don't filter
> >anything shorter than a /24.  While routing policy and allocation
> >policies don't need to necessarily match, it is not unreasonable to have
> >them in alignment.
> 
> I'm not doubting the claim but is there data to back it up?

It's a difficult thing to verify, frankly.  You can do your own
experiment if you have address space that's suitable for such things.
Have a look at http://www.arin.net/reference/ip_blocks.html#ipv4.  Pick
an aggregate in one of the /8s that isn't allocated down to /24s.
Announce a /24 out of a larger allocation, and start poking at
random route servers.

I did this experiment with a /24 at two different times spread over
about 10 months.  I found that my initial try had that /24 blocked
across two or three major ISPs.  My more recent run showed that only
one blocked that /24, and that was a bit suspect (it popped up in a few
route servers, so I suspect there was some weird one-off filtering
between a couple of the bigger folks out there...hard to say.)
Your mileage may vary.

In any event, I don't think there's definitive data, and few ISPs will
readily fess up to there exact filtering strategy.  It would be pretty
easy to filter based on the prefix lengths on that page mentioned
above...or just use /24.  The latter would *seem* to be the more common
strategy.

Like I said, difficult to verify, but reasonable to believe.

-David