Undersea Cables and the Speed of Light: Latency Limits

Quick answer: signals in optical fibre travel at about 200,000 km/s, which is roughly two thirds of the speed of light in a vacuum. That means each kilometre of fibre adds about 5 microseconds one way, or about 1 ms of round-trip time for every 100 km of cable. Most data crossing an ocean travels in submarine fibre-optic cables, and because those cables and the routes connecting to them are longer than a straight line, a server on the other side of the world cannot respond in under roughly 150 to 200 ms, however fast your connection or your computer is. Distance is the one part of latency that money cannot buy down.

How fast is light in fibre?

In a vacuum, light travels at about 299,792 km/s. In a material, it is slower by a factor called the refractive index. Optical fibre is made of glass with a refractive index of roughly 1.45 to 1.5, so light in it travels at about 299,792 ÷ 1.47, which is close to 204,000 km/s. Round figures of 200,000 km/s (200 km per millisecond) are used for planning, and that is what this guide uses.

Fibre distance	One-way delay	Round-trip delay (minimum)
1 km	5 µs	10 µs
100 km	0.5 ms	1 ms
1,000 km	5 ms	10 ms
5,000 km	25 ms	50 ms
10,000 km	50 ms	100 ms
20,000 km	100 ms	200 ms

The half of the Earth's circumference, about 20,000 km, is the longest great-circle distance between two points on Earth. A perfectly straight fibre between opposite sides of the planet would therefore give about 200 ms round trip. Real networks are never perfectly straight, so the worst-case floor for the most distant pairs is a little higher, and typical measured values between far-apart regions are often 250 to 350 ms.

How do undersea cables work?

A modern submarine cable is a bundle of optical fibres, typically several pairs, wrapped in copper for power, steel wire for strength and plastic insulation. In deep ocean it can be about as thick as a garden hose, while near shore it is armoured and often buried in the seabed to protect it from anchors and fishing gear. Each fibre pair carries data as pulses of laser light, and one pair can carry a very large amount of data by sending many wavelengths (colours) of light at once, a technique called wavelength-division multiplexing.

Light fades as it travels, so along the cable are repeaters (optical amplifiers), spaced every few tens of kilometres, which boost the light signal without converting it to electronics. They are powered from the shore through the cable's copper. Because the amplification is optical and done in the glass, repeaters add very little delay, so the latency of an undersea cable is almost purely propagation: length ÷ 200 km/ms. At each end the cable comes ashore at a landing station and connects to terrestrial networks, where routers and switches add small delays of their own.

Several hundred cable systems are in service, linking nearly every continent and carrying the large majority of international internet traffic. Satellites carry only a small fraction of it. Capacity is added regularly as new cables are laid, but the construction costs are high, and they follow paths that suit geography, politics and landing-station locations rather than the shortest line.

Why are real routes longer than the straight line?

Four factors lengthen the real path, and each one adds latency:

• Seabed and coastline. Cables avoid steep slopes, shipping lanes, fishing grounds, volcanic areas and earthquake zones, and must land where landing stations exist.
• Overland legs. Between a landing station and a data center, traffic follows railway, road or pipeline corridors, and fibre is rarely straight.
• Network topology. Traffic is handed between networks at specific meeting points, so a packet from one European city to another may pass through a hub in a third city. See routing, BGP and peering.
• Slack and spare length. Cables include extra length for repairs and for the shape of the seabed, and fibre inside cables is slightly longer than the cable itself because of how it is laid.

A useful rule is that real fibre paths are often 20 to 50 percent longer than the great-circle distance between two cities, and sometimes much more when the route detours. Add per-hop processing and access-network delay and the measured round-trip time for a city pair is commonly 1.3 to 2 times the straight-line physical minimum.

Typical minimum and real latency between regions

The table gives the straight-line distance, the absolute physical floor if fibre followed that line, and a typical range seen in practice. The ranges are approximate and change with routes and conditions.

City pair (approx.)	Straight-line distance	Physical floor RTT	Typical real RTT
London to Paris	about 340 km	about 3.4 ms	roughly 8 to 15 ms
London to Frankfurt	about 640 km	about 6.4 ms	roughly 12 to 22 ms
New York to Chicago	about 1,150 km	about 11.5 ms	roughly 18 to 30 ms
New York to London	about 5,570 km	about 56 ms	roughly 65 to 80 ms
New York to Los Angeles	about 3,940 km	about 39 ms	roughly 60 to 80 ms
Los Angeles to Tokyo	about 8,800 km	about 88 ms	roughly 100 to 140 ms
London to Singapore	about 10,850 km	about 108 ms	roughly 160 to 200 ms
Los Angeles to Sydney	about 12,050 km	about 120 ms	roughly 140 to 170 ms
London to Sydney	about 17,000 km	about 170 ms	roughly 250 to 300 ms

Notice how the real figures sit above the floor and how the gap grows when routes must detour. Traffic from Europe to Asia and Australia depends on a limited number of corridors, such as those through the Red Sea and Egypt, or around Africa, so the same city pair can differ by tens of milliseconds depending on which cable carries it on a given day.

What happens when a cable is cut?

Cable faults happen many times per year around the world, usually caused by ship anchors or fishing activity near shore, and sometimes by earthquakes or underwater landslides. Because networks have multiple routes, most users see no outage. Instead, traffic shifts to remaining cables, which are longer or already busy. Effects visible in latency tests include a step upward in round-trip time on affected routes (an extra 30, 60 or even 100 ms is not unusual if traffic moves to a much longer detour), more jitter and packet loss at busy times because the remaining capacity is shared, and gradual recovery after repair. Repair requires a specialised cable ship, which must travel to the site, lift the cable to the surface and splice it, so it can take days or weeks.

Can the speed-of-light limit be beaten?

For ordinary glass fibre the limit is set by physics. A few ways of reducing propagation delay exist, but each has limits:

• Shorter routes. New cables that take a more direct path, or terrestrial links that cut corners, save milliseconds. This is the most practical improvement.
• Hollow-core fibre. Light travels through air-filled channels, close to its vacuum speed, so propagation delay can fall by roughly a third compared with glass. It is deployed on limited routes, and long undersea use is not yet routine.
• Microwave and radio links. In air, radio signals travel at nearly the vacuum speed. Some financial firms use line-of-sight microwave towers between trading centres to beat fibre by a few milliseconds, but capacity is tiny and weather affects it.
• Low-orbit satellites with inter-satellite links. Because light in space is faster than light in glass, such systems can in principle have lower delay than fibre over very long distances, though real performance depends on constellation design and routing.

For everyday users none of these change the rule that distance sets the floor. Satellite internet is covered in Wi-Fi, Ethernet, fibre, mobile and satellite latency.

How to use this when you read your own results

When you run the free latency test from your own device, GlobeLatency times real HTTPS requests to public servers around the world. Compare each result with the distance, not with a single number. A destination in your own country at 20 ms is normal; a destination on another continent at 150 ms is also normal and is not a sign of a problem. The "World" figure beside each destination shows what other visitors measure for it, and your verdict is based on the ratio to that, so being "good" on a faraway destination simply means you are near the typical value for that destination. See reading your results for how to interpret the tiles, and global comparison explained for how the world average is used. Because the test goes through an HTTPS request, expect each number to be a little higher than raw ICMP ping: it includes DNS, TCP and TLS setup when a new connection is needed.

Worked examples

Example 1: Is 80 ms from New York to London bad?

The straight-line distance is about 5,570 km. Fibre floor: 5,570 ÷ 200 = 27.9 ms one way, 55.7 ms round trip. Real cable paths are around 6,000 to 6,500 km (call it 6,300 km), which gives 31.5 ms one way and 63 ms RTT. Add 5 to 15 ms for routers, last-mile and server, and a result of 70 to 80 ms is right where it should be. A result of 150 ms would be worth investigating, while 60 ms would require a particularly direct path.

Example 2: A website host choice

A game server in Sydney is 12,000 km from Los Angeles by straight line. Its physical floor is 120 ms RTT. A player in Los Angeles who measures 160 ms is seeing roughly 40 ms above the floor, which is typical overhead of a real route plus access network. Nothing can bring that player below 120 ms, so the only way to a better experience is a server nearer to them. See data centers, cloud regions and the edge.

Example 3: After a cable cut

A user in Mumbai normally measures 140 ms to a server in London. After a regional cable fault, traffic takes a longer path via a different corridor and the same test shows 230 ms with more variation. The extra 90 ms is roughly 9,000 km of additional fibre round trip (90 ms × 200 km/ms ÷ 2 = 9,000 km of extra path each way... in other words, about 4,500 km of extra one-way distance, since 90 ms RTT = 45 ms one way = 9,000 km one way). A rule of thumb check: extra RTT in ms × 50 gives extra one-way kilometres, so 90 × 50 = 4,500 km extra one way.

Common mistakes and myths

• "Upgrading to gigabit fibre will cut my ping to Asia." A faster plan changes bandwidth, not distance, so the 150+ ms floor to far destinations stays. See latency vs bandwidth vs throughput.
• Using the great-circle distance as the real distance. Real cables are longer. Use it only for the minimum.
• Expecting light in fibre to travel at 300,000 km/s. That is the vacuum speed. In glass it is about 200,000 km/s.
• Blaming the ISP for physics. If your measured latency is within roughly 30 to 50 percent of the floor for the distance, the network is doing about as well as can be expected.
• Believing a single big number means a fault. Compare similar distances and similar destinations, not different continents.
• Thinking the sea is the slow part. The cable itself is usually efficient; much of the extra delay comes from detours, handoffs and the access networks at each end.

What to do next

Estimate the floor for your own situation with the rule of thumb: distance in km ÷ 100 gives minimum RTT in ms, then multiply by 1.3 to 2 for a realistic range. Run the free latency test to a mix of nearby and far destinations and see whether your numbers follow distance. Then read what is a good latency to set expectations by use case, and troubleshoot high latency if a nearby destination is much slower than distance suggests. For a map-based view of how your connection reaches different parts of the world, open the live map.