Today's Internet topology generators can produce graphs that are
reasonably representative of network connectivity at either the level
of autonomous systems (ASes) or at the level of individual routers.
However, accurate connectivity models merely scratch the surface of
the topology generation problem.  For example, in order to conduct
meaningful router-level simulations, most applications require
annotations to the generated graphs such as link latencies, link
bandwidths, geographical coordinates and autonomous system
associations.

Unfortunately, our ability to accurately model these features is
currently impaired by our sharply limited understanding of them in the
Internet itself.  In this talk, we report on measurement studies we
have conducted to better understand the geography of the Internet.
Using worldwide router inventories from Skitter and Mercator, and
mapping tools from Ixia and Akamai, we map each router to its
geographical location.  We then study the relationships between router
location and population density; between geographic distance and link
density; and between the size and geographic extent of ASes.  We
conclude with our plans for incorporating the results of our findings
into our BRITE topology generation framework.