The mechanics of transoceanic packet routing

In a TikTok post, @charleszheng0 describes the global internet as a feat of engineering where a photo travels from a phone to a destination 4,000 meters below the Pacific Ocean in 200 milliseconds. The system relies on a distributed network of routers that pass data packets based on addresses rather than a predetermined path. This architecture allows for autonomous rerouting and high-speed transmission across transoceanic fiber optic cables.

How it works

Data transmission relies on a series of conversions between physical mediums and logical protocols. A device first translates visual data into a binary format, which is then encrypted to obscure the content from third parties. Because large files exceed the capacity of a single transmission window, the system breaks the data into smaller packets. Each packet carries a header containing a destination address and a sequence number. These packets move through various physical layers, changing from radio waves to light and back again, while routers perform the logic of moving them closer to the destination. The process ensures that even if a physical link fails, the network finds an alternative route to maintain connectivity.

1. Source DeviceThe phone converts the photo into 16 million ones and zeros and encrypts the data.
2. PacketisationThe system splits the data into 1,500 smaller packets with labels for addresses and sequence numbers.
3. Radio TransmissionThe phone sends the packets as radio waves to a nearby cell tower.
4. Fiber Optic ConversionThe cell tower converts radio waves into light to enter a fiber optic cable.
5. Router NetworkRouters read the address on each packet and pass it to the next machine in the chain.
6. Subsea CableFiber optic cables carry the light across the ocean floor, using repeaters to boost the signal every few dozen kilometres.
7. Reception and ReassemblyThe receiving phone identifies missing packets via sequence numbers, requests retransmission, and decrypts the final data.
8. DisplayThe phone puts the packets in order and displays the photo on the screen.

What it takes to run

Operating this infrastructure requires a global coordination of hardware and protocols. For a company to use this network, they must integrate with the local telecommunications providers that manage the cell towers and the first-mile fiber connections. The data must be prepared in a state where it can be segmented into packets, encrypted for security, and assigned specific headers for routing. Maintenance is not centralised. It depends on a distributed model where various entities manage different segments of the cable, the repeaters on the ocean floor, and the routers at exchange points. Reliability depends on the universal adoption of the protocols that allow these disparate machines to recognise and pass each other's packets.

• Cell towers and local radio frequency infrastructure
• Fiber optic cables and transoceanic repeaters
• Routers with logic to read headers and route packets
• Standardised protocols for packet sequencing and error correction
• Encryption keys for secure data reconstruction

Where it breaks

Distance and physical degradation impose hard limits on transmission. Signal attenuation in the glass fiber requires repeaters to boost light intensity every few dozen kilometres. If a repeater fails, the link becomes unusable. Latency is also a physical constant; light cannot travel faster than its medium allows. At the edges, congestion at router exchange points creates bottlenecks where packets queue, increasing the time between transmission and arrival. Encryption adds a computational overhead to the source and destination devices. While the network handles the transport, the end-to-end integrity depends on the hardware's ability to decrypt and reassemble fragments without errors. The system does not guarantee delivery; it only provides the mechanism for requesting missing data.

• Signal loss in the physical medium requires active amplification.
• Router congestion increases latency beyond the speed of light limits.
• Packet loss forces retransmission, doubling the time for specific data segments.
• Hardware limitations at the edge dictate the maximum encryption throughput.

How we would build it

We approach infrastructure by designing for the failure of individual components rather than assuming a perfect path. When building systems that rely on distributed networks, we treat every hop as a potential point of failure. This requires a shift from monolithic thinking to a strategy of graceful degradation and stateful reassembly. You need a system that can handle out-of-order delivery and packet loss as a standard operational condition, not an exception. We focus on the logic that governs how data is segmented, validated, and reconstructed at the destination.

1. Packet SegmentationWe split large payloads into smaller units to ensure they can traverse diverse network paths without being dropped due to size limits.
2. Header AssignmentEach segment receives a unique identifier and destination address to allow routers to make independent forwarding decisions.
3. Asynchronous ReassemblyThe receiving client buffers incoming data and uses sequence numbers to reconstruct the original object regardless of arrival order.
4. Retry LogicWe implement a mechanism for the client to request specific missing segments from the source to ensure data integrity.

Engineers who succeed build for failure by designing idempotent operations and state machines that reconcile divergent logs. They do not treat a lost packet as a bug to be fixed, but as a constant input to be handled.