The 30 Millisecond Rule: Why Distance Still Sets the Speed of Every Packet

Archi4 min read

networkinglatencytelecom

Contents
  1. The Number That Fiber Cannot Beat
  2. Where the Milliseconds Go
  3. The speed of light inside the box
  4. Why Content Networks Moved Closer
  5. The physics does not negotiate
  6. What This Means for Design
  7. The four millisecond trade
  8. A Useful Rule of Thumb

The Number That Fiber Cannot Beat#

A trading firm in Chicago pays for a microwave link to New Jersey for one reason: it saves roughly four milliseconds over fiber. Four thousandths of a second. That premium exists because light in glass travels at about two-thirds of its speed in a vacuum, roughly 200,000 kilometers per second. Every 200 kilometers of fiber adds a millisecond of one-way delay. At internet scale, that arithmetic sets a hard floor under every application you run.

Engineers call this propagation delay, and it is the one part of latency that no amount of money, code optimization or newer router can remove. You can buy better hardware. You cannot buy more speed of light.

Where the Milliseconds Go#

Round trips are what users feel. A request from London to a server in Virginia covers about 5,900 kilometers each way. On a straight line at 200,000 km/s that is about 29.5 milliseconds one way, so nearly 60 milliseconds before the server has done anything at all. Real routes bend around coastlines and hop through switching centers, so the practical figure is usually higher, often 70 to 80 milliseconds.

That is before you count the other delays in the chain. Serialization time, the period a router needs to push bits onto a link, is small on modern gear but adds up across many hops. Queuing delay appears whenever a link runs hot. Processing delay comes from every device that inspects a packet header. On a congested path, queuing can dwarf propagation. On a clean long-haul path, distance is the bill.

This is why a page that feels instant on a laptop in Frankfurt can feel sluggish on the same connection type in Singapore. The software did not change. The geography did.

The speed of light inside the box#

Latency also starts inside the machine. A request leaving a browser passes through the operating system's network stack, the network interface, a switch, a router, and often several middleboxes before it reaches open fiber. Each hop costs microseconds to milliseconds. On a local network these costs dominate. Across an ocean, they are rounding errors next to propagation.

Why Content Networks Moved Closer#

Content delivery networks exist largely to shrink the distance term. If a copy of a file sits 50 kilometers from the user instead of 5,000, the round trip falls from tens of milliseconds to a handful. That single move explains a large share of the perceived speed of the modern web, more than any single protocol revision managed on its own.

The same logic shapes cloud regions. Providers now offer small edge locations in secondary cities precisely because a 20 millisecond advantage is enough to change how an application feels. Gaming companies place servers by population density, not by data center economics. Voice and video platforms route calls through the nearest relay they can find.

The physics does not negotiate#

Some vendors market products as if latency were a software bug. It is not. A satellite in low Earth orbit at 550 kilometers adds roughly 2 milliseconds one way just for the trip up and down, plus routing overhead, which is why low-orbit constellations beat geostationary satellites by hundreds of milliseconds. Geostationary orbit sits about 35,786 kilometers out, so the round trip alone is close to 240 milliseconds of pure physics before any equipment adds its share. No ground station upgrade changes that number.

What This Means for Design#

If distance sets the floor, the practical response is to stop pretending it does not exist. Three patterns keep showing up in systems that handle it well. , Move computation toward users. Caching, regional deployments and edge functions all trade complexity for milliseconds., Reduce round trips. A protocol that needs six back-and-forth exchanges pays the distance cost six times. Batching requests and reusing connections cuts the bill., Set expectations from physics. A support team arguing about a 90 millisecond response time between two continents should first check the map, then the code.

The four millisecond trade#

Return to the Chicago firm. Microwave links beat fiber over that route because radio travels through air at nearly the speed of light in a vacuum, while glass slows light by about a third. The saving is real but modest, and the link is fragile in weather. The firm accepts that trade because in its market four milliseconds is worth more than reliability engineering. Most organizations would not make the same call, and that is fine. The number still explains the behavior.

A Useful Rule of Thumb#

For rough planning, assume 5 microseconds per kilometer of fiber, one way. A 1,000 kilometer link costs about 5 milliseconds each way, 10 for a round trip. Add a few milliseconds for equipment and routing, then compare that figure to whatever latency target your application claims.

If the target is below the physical floor, the target is wrong. Fix the architecture or move the server. The speed of light is not a configuration setting, and the 30 milliseconds between two distant cities will still be there after the next hardware refresh, the next protocol draft and the next cloud price cut. That is not a limitation to solve. It is the map drawing itself into the network.

More articles