Abstract |
The Internet is getting better and better at delivering content to end-users; this shift is
spearheaded by Internet giants such as Google and Facebook. Nevertheless, a number of
applications (e.g., peer-to-peer, Blockchain) rely on user-to-user connections, which
raises the –still open– question of how end-users are connected with each other. This is
further stressed by the fact that multiple suspicious incidents of path manipulation for
user- to-user communications have been reported.
In this work, we use active measurements (i.e., traceroutes between RIPE Atlas vantage
points) and publicly available datasets to explore the interconnectivity of the user-facing
networks with the largest user populations in any given country. We combine user
population per autonomous system (AS) estimates from APNIC with data plane
measurements and provide insights into the user-to-user connectivity for 114 countries,
over time. In order to study per-country interconnectivity, we construct a framework that
stores and processes massive traceroute datasets, making refined results available via an
online public API. On a monthly basis, we analyze ∼420K traceroute paths, from ∼3,5K
RIPE Atlas probes in ∼2,6K ASes.
We derive statistics and comparisons between countries in terms of: (i) out-of-country vs.
in-country paths, (ii) direct connections vs. intermediary networks, (iii) IXP crossings vs.
non-IXP crossing paths. We discover among other findings that over time 20% to 50% of
the user-to-user connections in Greece cross an IXP; while in the U.S., the fraction of such
connections is only ∼3%. We also propose a methodology to infer the transit
betweenness of networks in the user-to-user paths. For example, in the U.S. 8% of the
user-to-user connections flow through the incumbent provider, consistently in time.
Besides, we go beyond eyeball networks, and focus on the differences between the two
Internet protocols (IPv4/IPv6) in terms of path lengths, paths staying in or going out of a
country, as well as IXP crossings. We observe paths in IPv6 to be shorter than IPv4 for
almost all countries. Moreover, in the U.S. the fraction of paths that cross an IXP in IPv6
is ∼50%; two times more than in IPv4. Finally, we evaluate the coverage of RIPE Atlas on
user populations around the globe.
|