Your line, graded against the speed of light
68%
is the most a line of glass can reach
Light in optical fibre runs at 1/1.4682 of its speed in a vacuum, so a route made only of glass can never pass 68.1 % of light.
Timing your line runs in your browser and needs JavaScript. The floors between 79 cities are on plain pages: Floors by City.
From your connection's city
Our host sees your IP address, as every host does. The page asks it for your city to compute distances, and keeps nothing. Privacy
Your Regions
32 cloud regions, timed one at a time
Each region is timed with one warm-up request and five timed ones. Best is the fastest of the five and sets the grade; median is the middle one. Floors are round trips, there and back along the great circle.
| Region | Grade | Your best | Your median | Vacuum floor | Glass floor | Distance |
|---|---|---|---|---|---|---|
| The table fills in as each region answers. | ||||||
Where the Time Goes
Light, glass, and everything else the route adds
Each best round trip, taken apart. Light is the vacuum floor. Glass is what fibre adds over vacuum for the same distance. Everything else is the rest: cables that leave the great circle, routers, queues, and the browser itself.
The Straight Wire
Chicago to New Jersey, bought by the millisecond
In August 2010 Spread Networks opened a fibre route between Chicago and the New Jersey data centres, laid straighter than any before it: 825 miles of glass, quoted at 13.33 ms round trip. Media reports put its cost at about US$300 million. Its quoted one-way time, 6.65 ms, grades at 57 % of light against the 3.78 ms floor from central Chicago to Carteret, New Jersey, where Nasdaq's matching engine ran. Glass alone stops at 68 %.
Within two years, trading firms had licensed microwave links on the same route, because air carries light at nearly its speed in a vacuum. Gregory Laughlin, Anthony Aguirre and Joseph Grundfest estimated those links at 4.20 to 5.20 ms one way: 90 % of light for the fastest, 73 % for the slowest. No fibre could have got there.
- Light in a vacuum, the floor3.78 ms
- Microwave, fastest estimate (2012)4.20 ms
- Light in glass, the glass floor5.55 ms
- Spread Networks fibre, quoted (2010)6.65 ms
- S. Hardy, "Spread Networks to debut low latency New York to Chicago dark fiber route", Lightwave, 22 June 2010: 825 fiber miles, 13.33 ms.
- G. Laughlin, A. Aguirre and J. Grundfest, "Information Transmission Between Financial Markets in Chicago and New York", The Financial Review 49 (2014) 283–312: Spread Networks at about 6.65 ms one way; the cost of about $300 million from media reports (C. Steiner, "Wall Street's Speed War", Forbes, 27 September 2010); microwave links at 4.2 to 5.2 ms; the 3.93 ms floor from Aurora, Illinois, which this page's engine computes as 3.96 ms from the GeoNames points of Aurora and Carteret.
The Working
Every floor computed here, every figure sourced
How is the line timed?
Your browser asks each region's public endpoint, dynamodb.<region>.amazonaws.com/ping, for a few bytes, one region at a time, so no request queues behind another. The first request opens the connection (the name lookup and the two handshakes) and is thrown away. The next five are timed with the browser's own clock. The fastest is the best your line can do at that moment, and the grade uses it.
What is the floor?
Light in a vacuum covers 299,792.458 km each second, exactly, by the definition of the metre. The floor is the great-circle distance between you and the region on the WGS84 ellipsoid, there and back, at that speed. Optical fibre slows light by its group index, 1.4682 for Corning's SMF-28e+ at 1550 nm, so the glass floor is 1.4682 times longer, and a route made only of fibre can never pass 68.1 % of light. A grade above that means part of the path ran through air, or your location is off.
What is the grade?
The vacuum floor's round trip divided by your fastest measured round trip, to the nearest whole percent. A line at 50 % takes twice as long as light would.
What can the page not see?
- AWS names its regions for cities, states and countries, never for buildings. Each region named for a city sits at that city's GeoNames point, good to ±0.33 ms on the round trip. These are named for a state or a country, and sit at a stand-in city with a wider radius:
- N. Virginia (us-east-1) at Ashburn, ±0.40 ms
- Ohio (us-east-2) at Columbus, ±1.33 ms
- N. California (us-west-1) at San Jose, ±1.67 ms
- Oregon (us-west-2) at Bend, ±2.00 ms
- Montreal (ca-central-1) at Montréal, ±0.67 ms
- Querétaro (mx-central-1) at Santiago de Querétaro, ±0.67 ms
- Dublin (eu-west-1) at Dublin, ±0.67 ms
- Spain (eu-south-2) at Zaragoza, ±1.00 ms
- UAE (me-central-1) at Dubai, ±1.00 ms
- Malaysia (ap-southeast-5) at Kuala Lumpur, ±0.67 ms
- Thailand (ap-southeast-7) at Bangkok, ±0.67 ms
- The browser adds its own time to every sample, a fraction of a millisecond the page cannot subtract, so every grade reads a little low.
- Your connection's city comes from your IP address. Behind a VPN or a mobile carrier it can be hundreds of kilometres off, and then a grade can pass 100 %. Pick your city and the floors follow.
- The great circle is not the deepest floor: a straight line through the Earth is shorter. Every cable anyone can lay follows the surface, so the surface is the floor that matters.
- AWS's Bahrain region (me-south-1) does not answer from a browser, so it is left out.
How are the numbers checked?
Distances match GeographicLib on the WGS84 ellipsoid to within 100 m on ten city pairs, against a tolerance of 0.5 %. The floors match a hand computation to 1 µs. The Aurora to Carteret floor lands within 0.05 ms of the published 3.93 ms. The receipts run with every build.
Save the Map
Your routes, lit by grade, at print size
The map saves once the first region is graded: an SVG at 594 × 420 mm (A2) and a 4096-pixel PNG.