How the Internet Routes Data Between Continents in Milliseconds

Light Through Glass, Crossing an Ocean

A message sent from Tokyo to London travels roughly 9,600 kilometers of undersea fiber-optic cable and arrives in approximately 95 milliseconds — about the time it takes to blink. That speed depends on light pulses moving through glass at roughly two-thirds the speed of light in a vacuum, guided by a physical and logical infrastructure built over six decades. Understanding how that journey happens requires examining both the cables under the ocean and the routing protocols that decide which path each packet takes.

The Physical Layer: Submarine Cables

The internet's backbone is largely underwater. As of 2025, over 574 submarine cable systems span approximately 1.4 million kilometers, carrying more than 99% of international internet traffic. Satellite connections handle less than 1% of intercontinental data — high latency and limited bandwidth make them unsuitable for most applications at scale.

Modern submarine cables use dense wavelength-division multiplexing (DWDM), which encodes data on dozens of separate light wavelengths simultaneously within a single fiber strand. A cable system like Google's Dunant, which entered service in 2021, carries around 250 terabits per second across the Atlantic — enough to transfer the entire Library of Congress every second.

Cable landing stations — shore-based facilities where submarine cables connect to terrestrial networks — act as the handoff points. From there, terrestrial fiber carries data to internet exchange points (IXPs) in major cities.

Autonomous Systems and BGP

The internet is not a single network. It is approximately 75,000 independently operated networks called Autonomous Systems (ASes), each identified by a unique ASN (Autonomous System Number). Every major ISP, cloud provider, university, and corporate network operates one or more ASes.

These ASes exchange routing information using the Border Gateway Protocol (BGP), invented by engineers Kirk Lougheed and Yakov Rekhter on cocktail napkins in 1989 and still the foundation of internet routing today. BGP is a path-vector protocol: each AS advertises which IP address prefixes it can reach and the sequence of ASes a packet would traverse to get there. Routers use these advertisements to build a routing table — a lookup structure that maps destination addresses to outbound interfaces.

How a Route Is Chosen

BGP does not optimize purely for speed or shortest path. It selects routes based on a hierarchy of policy attributes in this order:

• Local preference: Set internally; higher value preferred. Used to prefer one upstream provider over another.
• AS path length: Shorter AS paths preferred when local preference is equal.
• Origin type: IGP > EGP > incomplete (legacy attribute, rarely decisive).
• MED (Multi-Exit Discriminator): Hint from a neighbor about preferred entry point into their network.
• eBGP over iBGP: Routes learned from external peers preferred over internal peers.
• IGP metric: For tie-breaking, the lowest internal cost to reach the BGP next-hop wins.

Because policy governs route selection, the fastest physical path is not always chosen. A packet from New York to Amsterdam might route through London if a carrier has a peering agreement that makes that path more economical, adding 10–15ms of unnecessary latency.

How a Packet Makes the Trip

Sending an HTTP request triggers a multi-layer process. The browser assembles data into TCP segments, which are encapsulated in IP packets with source and destination addresses. Each packet is typically 1,500 bytes (the standard Ethernet MTU). A 1MB image becomes roughly 700 separate packets, each navigating independently.

At every router along the path, the device reads the destination IP address, performs a longest-prefix match against its routing table, and forwards the packet to the next hop — a process that takes nanoseconds on modern hardware. A transcontinental route might traverse 15–25 router hops. The traceroute command makes this visible: running traceroute google.com from New York to a Google server in Europe reveals each hop, its location, and the round-trip time added.

Latency: Physics Sets the Floor

No engineering can overcome the speed-of-light limit. Light travels through fiber at approximately 200,000 km/s (slower than in a vacuum due to the glass refractive index). This sets a theoretical minimum latency for any given route.

The gap between physical minimum and real-world latency reflects queuing delays at routers, protocol overhead, and suboptimal routing paths caused by BGP policy decisions.

Internet Exchange Points

Internet Exchange Points (IXPs) are facilities where multiple networks physically interconnect, allowing traffic to pass directly between them without traversing a third-party transit network. The DE-CIX in Frankfurt is the world's largest, regularly handling over 14 terabits of peak traffic. AMS-IX in Amsterdam and LINX in London are other major hubs. By peering at an IXP, two networks can exchange traffic without paying a transit carrier, reducing both cost and latency.

Resilience and Redundancy

Cables break. Ships drag anchors, earthquakes shift seabeds, and sharks occasionally bite cables (a real documented phenomenon). Resilience comes from redundancy: multiple cable paths between major regions ensure that traffic reroutes automatically when one cable fails, typically within seconds via BGP reconvergence. The 2022 Tonga volcanic eruption severed the island's sole submarine cable link, illustrating what happens when redundancy does not exist — the country lost almost all connectivity for five weeks.

The infrastructure that moves a text message across an ocean in milliseconds represents one of the most complex engineered systems on Earth — and one that most people interact with hundreds of times daily without ever thinking about what happens between tap and response.