How the internet actually works, and why nobody is in charge of it

Maneshwar is developing LiveReview, an AI code review tool for business-critical systems. When opening a video, it begins playing within a second as your request travels through fifteen companies, possibly crossing an ocean. Contrary to explanations, the internet is not a single network but 75,000 independent networks that mutually agree on traffic handovers. Your ISP, university, and Cloudflare are examples of these networks.

There are three parts to this arrangement: the edge, access network, and core. End systems, like your phone, laptop, or server, make up the edge and communicate through access networks like home fiber, office networks, or 5G. The core consists of routers and interconnecting links, with no central control or ownership.

The design is clever: data is divided into packets stamped with origin and destination, then sent individually to find their own path. Packets can take various routes, arrive out of order, or be lost. This might seem inefficient but offers efficiency, damage resilience, and scalability. The end-to-end argument states that routers simply pass packets, while reassembly and retries occur at the endpoints.

Each data blob is wrapped four times with layers: application, transport, network, and link. Decapsulation happens at the receiving end. This modular design allows changing physical networks without altering applications. Traceroute shows each router along the path, owned by real companies, choosing to forward packets based on routing decisions made before the packet leaves the local network.

BGP, the Border Gateway Protocol, shares reachability information between networks. However, BGP assumes network operators are honest and competent, lacking built-in mechanisms to verify the truth of announcements.