... routers¹

Actually, router interfaces are discovered, not routers.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... extent²

We could have used a diverse set of public traceroute servers [22] to overcome this limitation. However, the large volume of traceroutes that we were looking to run from each source precluded this.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... routers³

To be precise, DNS names are associated with router interfaces, not routers themselves. However, for ease of exposition we simply use the term router.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... BBNPlanet⁴

BBNPlanet is now called Genuity, but the router names are still in the bbnplanet.net domain.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... location.⁵

Of course, it is possible to include psc and cmu as codes. However, we refrain from doing so since we only want to include those codes in GeoTrack that inherently indicate location. Doing otherwise would lead us down the path of exhaustive tabulation, which is undesirable.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... increases.⁶

The jaggedness of the curves arises because of the large variance in distance ratio for small values of geographic distance. The 5th and 95th percentile marks for the 100-400 km bucket are (1.00,20.50) for the end-to-end case and (1.00,4.22) for the intra-ISP case. The corresponding marks for the 4000-4400 km bucket are (1.01,1.57) for the end-to-end case and (1.00,1.18) for the intra-ISP case.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... 2).⁷

While the choice of these thresholds is arbitrary, they capture the intuitive notion of circuitous and non-circuitous routes. Note that there may be paths that do not fall into either category.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.