The second Russian hackathon on Internet measurements
Original goal: measure latency between geographically close points and pinpoint abnormally high latencies (bad peering?) compared to speed of light in fiber.
The goal was shifted to replace active measurements with historical measurements analytics, modified goal was to find latency outliers between two geocoded endpoints, especially violations of speed of light (showing geocoding incorrectness).
Assumptions:
- All IP addresses within /24 are quite close geographically
- Geolocation data for RIPE Atlas probes from YYYYMMDD.json may be ultimately trusted and is almost static
So far these assumptions turned out to be non 100%-true.
Probe #27782 and probe #6254 are within same /29 but they're 350 km away, so assumption of near-zero latency between these two hosts is wrong and calculations assuming near-zero latency between hosts in same /24 lead to contradiction with speed of light being limited.
Also, some probes have their "latency-based" location quite far from "declared" location. Some of these probes are "travelling", some estimates of latency-based location may be wrong due to wrongly located destination locations (that are based on /24 locality assumption) and so on, some of these probes can be seen in fishy.ipynb.
Latency measurement may be used to find outliers in geolocation, but common factor of ~2x difference between actual RTT and RTT estimated with speed of light in fiber and traversed distance makes it not so precise.