Check out the new USENIX Web site. next up previous
Next: Temporal properties of routing Up: Effect of geographic location Previous: Multiple sources in different


U.S. versus Europe

We now analyze the distance ratios for paths in Europe and compare these to the distance ratios for paths in the U.S. We consider paths from the 17 U.S. sources to destinations in the LibWeb set and also paths from the 3 European sources to destinations in the EuroWeb set. Thus, all of these paths are contained either entirely within the U.S. or entirely within Europe. We do not consider paths from U.S. sources to European destinations (or vice versa) because the distance ratio for such paths tends to be dominated by long transatlantic links (which tends to push the ratio towards 1).

Figure 6: CDF of distance ratio for paths within the U.S. and those within Europe.
\begin{figure}
\centerline{\psfig{file=figs/USvsEurope.eps,height=3.0in,width=4.5in,angle=0}}
\vspace{-0.20in}
\textbf{}
\end{figure}

In Figure 6, we show the distribution of the distance ratio for three sources: Berkeley in the U.S., and Stockholm (Sweden) and Bologna (Italy) in Europe. We observe that the distance ratio tends to be larger for the European sources compared to Berkeley, especially in the tail of the distribution. We attribute this to three causes. First, paths in Europe tend to traverse multiple regional or national ISPs. The complex peering relationships between these ISPs often results in convoluted paths. For instance, a path from Bologna to a host in Salzburg, Austria traverses 3 ISPs - GARR (Italian Academic and Research Network), Eqip/Infonet, and KPNQwest (a leading pan-European ISP based in the Netherlands) - and passes through Milan (Italy), Geneva (Switzerland), Paris (France), Amsterdam (Netherlands), Frankfurt (Germany), and Vienna (Austria). The linearized distance of the path is 2506 km whereas the geographic distance between Bologna and Salzburg is only 383 km. Second, in some cases the path from a European source to a European destination passes through nodes in the U.S. For instance, a path from Stockholm (Sweden) to Zagreb (Croatia) passes through a node in New York City belonging to Teleglobe, a large international ISP. In the event that the ISPs in Europe have better connectivity to ISPs in U.S., it would be appropriate for them to route their traffic through U.S. though the route may be more circuitous. Third, geographic distances in Europe tend to be smaller than the ones in U.S. As in the case of St Louis in Section 4.2.1, small detours in routing can be particularly expensive in terms of the distance ratio for paths between end-hosts in Europe.
next up previous
Next: Temporal properties of routing Up: Effect of geographic location Previous: Multiple sources in different
Lakshminarayanan Subramanian 2002-04-14