... has anybody got a link to the actual paper? Based on the press reports, the methodology seems flawed, and the claims about unprecedented scope look bogus.
The methodology is very flawed because it does not account for private internetworks, which do not exchange traffic with the Internet. Also, although they took some precautions to reduce the loss of their probe packets, there are still some things like ICMP blocking, which make large chunks of address space completely invisible to them. You can read their paper at <http://www.isi.edu/~johnh/PAPERS/Heidemann08c.pdf> It is interesting work from a technical standpoint, but from a policy standpoint, it is not terribly useful since it is intended to measure the public Internet, not the approved usage of the IPv4 address space. Remember, we approve the use of IPv4 addresses that are not assigned to hosts. For instance, a company can assign a /29 to a subnet with 5 hosts, and their LIR will count all 8 addresses as being in use, in conformance with RIPE policy. --Michael Dillon