What Is The Internet? (Considering Partial Connectivity)

The Internet was originally defined as "a collection of inter-connected networks". While this definition helps us understand what the Internet is, it is silent on when the Internet is not. We provide a testable definition of the Internet to clarify where the Internet "ends": disconnection when a country or an ISP secedes; persistent partial connectivity when major ISPs refuse to exchange traffic, isolating their customers; clarifying corner cases around carrier-grade NAT, unrouted public IP addresses, and interpreting conflicting observations from systems that detect Internet outages. Our definition identifies peninsulas of persistent, partial connectivity, and clarifies that outages are islands, with internal connectivity that is partitioned from the main Internet. Our definition is conceptual, defining an ideal asymptote of connectivity, but it enables new algorithms that provide an operational estimate of the number of size of peninsulas and islands. We use these algorithms to reinterpret data from two existing measurement systems, one covering 5 million /24 IPv4 networks and the other with 10k observers. A key result is that peninsulas are about as common as outages, newly clarifying the importance of this long-observed problem. We examine root causes, showing that most peninsula events (45%) are transient routing problems, but a few long-lived peninsulas events (7%) account for 90% of all peninsula time, suggesting country- or AS-level policy choices that last weeks or more. Finally, our definition confirms the international nature of internet: no single country can unilaterally claim to be "the Internet", but countries can chose to leave. With islands and peninsulas, our definition helps clarify the spectrum from partial reachability to outages in prior work.