Abstract
It's been more than 25 years since today's dominant Internet Protocol (IPv4) was deployed, yet today, in some ways, we know less about the Internet than we did in the 1980s. Seemingly simple questions today have at best poor estimates: how many hosts are there? how much is dynamically allocated, and where? how effectively are we using the IPv4 space? how does the cost of better management compare to changing to IPv6's huge space? These questions are increasingly important given predictions that ICANN will allocate the last unallocated addresses by 2011, that spammers exploit dynamic addresses, and that we just should know how big the net is.

To explore these questions we have taken multiple *Internet censuses*, attempting to contact (ping) all allocated IPv4 addresses every few months; and *Internet surveys*, probing about 1% of the Internet frequently for a week at a time. This talk will describe our new methodology. Firewalls are one of several clear sources of error in our approach; we will present data to suggest that, although far from perfect, we can estimate error. Finally we will use this new source of data to to begin answering some of the open questions listed above.

(This work is a joint effort with Joseph Bannister, Genevieve Bartlett, Xue Cai, Ramesh Govindan, Christos Papadopoulos, Yuri Pradkin at USC, Colorado State U., and Aerospace Corp.)