Evidence of spatial embedding in the IPv4 router-level Internet network

TL;DR

This paper provides evidence that the IPv4 router-level Internet network exhibits spatial embedding, with geographic attributes strongly correlating with network topology, which has implications for modeling and geolocation inference.

Contribution

The study introduces node subgraph summary statistics to demonstrate spatial embedding in the IPv4 router network, linking network structure to geographic attributes.

Findings

Router network shows strong correlation with geographic location.

Spatial embedding supports gravity models of Internet structure.

Potential for improved geolocation inference methods.

Abstract

Much interest has been taken in understanding the global routing structure of the Internet, both to model and protect the current structures and to modify the structure to improve resilience. These studies rely on trace-routes and algorithmic inference to resolve individual IP addresses into connected routers, yielding a network of routers. Using WHOIS registries, parsing of DNS registries, as well as simple latency-based triangulation, these routers can often be geolocated to at least their country of origin, if not specific regions. In this work, we use node subgraph summary statistics to present evidence that the router-level (IPv4) network is spatially embedded, with the similarity (or dissimilarity) of a node from it's neighbor strongly correlating with the attributes of other routers residing in the same country or region. We discuss these results in context of the recently proposed gravity models of the Internet, as well as the potential application to geolocation inferrence.