Coverage of GeoTrack

Of the 11,296 .net router names in our traceroute data set, 7842 were recognizable (approximately 70%). We compiled a list of 13 major ISPs with nationwide backbones in the U.S. or with international coverage: Sprintlink, AT&T, Cable and Wireless, Internet2, Verio, BBNPlanet⁴, Qwest, Level3, Exodus, PSINet, UUNET/Alter.net, VBNS, and Global Crossing. We found that 5,966 of the 6,859 router names for these major ISPs were recognizable (87%). In some individual cases, such as AT&T and UUNET, the recognizability was in excess of 95%.

Figure 2: The recognizability of router names as a function of the position of the router in the end-to-end path. The position is quantified by dividing the number of hops leading up to the router by the total number of hops end-to-end.

By manual inspection, we found that a large chunk of the router names which are unrecognizable by our tool have no meaningful codes to decipher their locations. Many unrecognizable router names tend to be concentrated in regional or campus networks. (For example, cmu.psc.net is a node in Pittsburgh, PA. However, since it does not contain a valid city or airport code, GeoTrack is unable to recognize its location.⁵) Figure 2 shows that recognizability is lowest close to the start and the end of the path. (The peak corresponding to the very beginning of the path is due to the source location always being known.) Thus most of the unrecognizable nodes are typically located in the vicinity of the source or the destination, so the resulting error in linearized distance is minimal. In the case of the 1995 data set, GeoTrack is able to recognize 1,289 out of 1,531 router names (approximately 84%). Interestingly, we noticed a huge difference in the naming convention used in 1995 and 2000. Hence we needed to create a new set of codes for the 1995 data set.