3 ms·
On the Suitability of RTT Measurements for Geolocation
- wongarsu 9y agoI wonder if this becomes more feasible if instead of drawing exclusion circles based on speed of light in fiber, you would use a graph of network links. For example the location and length of submarine cables is well known, and there is also somewhat decent information on the location and speed of other backbone links. If a RTT is 500ms, you can walk this graph of known links up to a distance of 250ms, and draw circles around every branching point equivalent to the remaining RTT budget in order to account for unknown links to consumers. That way you end up with a bunch of circles that combined contain all possible locations of the target. You could even take it a step further with empirical measurements of real delays and variance (based on time of day) between known locations to map unmapped parts of the network and to make educated guesses about the used links based on experienced RTT variance. I guess it's still hard to get better than an IP database unless the target uses top-notch networking equipment, given that a single ms means 300km difference. But maybe variance of RTT could give enough clues to rule out multiple links, giving you a way to enhance the location estimate from the IP database in some cases.
- mjevans 9y agoYou're largely correct in how you define the probable zone of inclusion. However as latency can always be artificially increased but never decreased, this becomes less useful for excluding 'nearer' entities.
- aidenn0 9y agoIt seems to me that an empirical exclusion based method could be used; there appears to be very few points that are close to the green line in their chart. Alternatively, one could use statistical multilateration. If you have more than 3 points, then you can draw multiple spheres and find an area that has a high likelyhood of being correct.