Innovation Blog: Much faster than light
The speed of light is constant throughout the universe, except in optical fibres. But users are willing to pay more for speed, writes Alan Burkitt-Gray
What is the speed of light? As anyone who’s studied physics will recall, Albert Einstein said in 1905 it’s constant throughout the universe – a velocity now measured at 299,792km/s. Engineers, who like to round numbers off, think of it as 300,000km/s, which makes it possible to do wavelength/frequency calculations in your head – a radio signal with a frequency of 100MHz, right there on your FM dial, has a wavelength of 3m.
Of course, I’ve assumed there that you know light and radio waves are different versions of the same thing: different frequencies, different wavelengths.
Red light has a wavelength of 650 nanometres which means it has a frequency of 462 terahertz. And red light, along with gamma rays, X-rays, all forms of light, right down to short-wave signals beloved by amateur radio enthusiasts, all travel at the same speed: (almost) 300,000km/s.
But things aren’t as simple as that. They never are. That figure is for the speed of light in a vacuum, in outer space, in most of the universe. If you’re a satellite person, that works well.
Even if you’re worried about radio signals in the atmosphere, between your mobile phone and the nearest mast, or from your nearest TV transmitter to your rooftop antenna, 300,000km/s holds good. In fact, the speed of light in air is 99.97% of the speed in a vacuum and, as we’ve already rounded that up from 299,792km/s, frankly, who cares?
Fast, but not that fast
But as telecoms people, the speed of light you’re most interested in is its velocity through glass, in a fibre, buried under the streets or under the sea. This is about 204,500km/s, only 68% of the speed in a vacuum, or a bit over two-thirds. Of course, the exact figure depends on the type of glass, but it’s still fast.
So let’s put that in terms you encounter in your everyday life. Light will travel about 200km through optical fibre in a millisecond or over 1,000km in 5ms.
As Einstein said, you can’t go faster than light. But here we’re not talking about light travelling through a vacuum but high-quality optical glass, and it’s fundamental to the operation of fibre that it conducts light more slowly than a vacuum does. It’s down to the refractive index of the glass – how much light bounces around inside the fibre – or “total internal reflection”, as physicists call it.
So, if you’re 2,000km from Ashburn, Virginia (let’s say you’re in Austin, Texas) then you’ll never get a signal to run between those two places in less than 10ms. Want lower latency? Can’t be done. And that’s ignoring the network electronics at each end. Even without all that electronic processing, 10ms one-way or a 20ms round-trip time is the best you can get, according to the laws of physics.
The weird stuff
And that’s why some telecoms people have tried weird stuff over the years to squeeze a few milliseconds out of the latency. One way is to go back to pure radio, because radio waves go through the atmosphere at pretty close to the speed of light – a good deal faster than light goes through fibre.
The trouble is – as all those high-frequency traders find – you need radio waves that can go over the horizon to send a signal 2,000km from Ashburn to Austin. That means heading to the short waves, radio with wavelengths of 20-60m. As anyone who’s every consorted with radio amateurs will tell you (as I did in my student days, a long time ago), short waves don’t work well during the day. They’re best at night.
And having frequencies of 5-15MHz means those signals can’t carry much data.
The other weirdness some telecoms engineers have tried is to make fibre routes as short as possible. There was a company called Spread Networks that decided to build, as near as was possible, in a straight line between the Chicago Mercantile Exchange and the Nasdaq data centre in New Jersey, a distance of 1,331km – giving a built-in, unavoidable latency of about 6.6ms.
Going the long way round would have taken the distance to around 1,500km, or 7.5ms latency. And that 0.9ms is financial death for high-speed traders.
Spread Networks spent $300 million on its project. A few years later, Zayo bought it for just $127 million. So that went well.
Save a picosecond or two
But the wackiest ideas come from those traders – or their IT managers – who insist that fibre in the data centre itself has to be as short as possible, even if it means shaving just 30cm off it.
Do a quick calculation: it takes light 1.5 picoseconds to travel 30cm along fibre. A picosecond is one-trillionth of a second. The difference in scale between 1.5ps and 1.5s is equivalent to the difference between 1.5s and 47,564 years. How important is that to high-frequency trading?
Here physicists are riding to the rescue with an alternative that’s been gathering a few enthusiastic followers: hollowcore fibre. In the early days of optical fibres, the usual press-release description was that they were “hair-thin strands of glass”. Hollowcore fibres, by contrast, are hair-thin hosepipes of glass.
The hole in the middle
The idea behind hollowcore is that light will travel down a hole in the middle a glass pipe, bouncing off its internal walls.
According to the University of Southampton, where the technology (pictured) has been developed, this means signals will travel at 99.7% of the speed of light.
If you put a kilometre of hollowcore fibre alongside a kilometre of conventional fibre, the signal through the hollowcore will arrive about 1.5 microseconds earlier. And for high-speed traders, 1.5µs is important. I mean, really important.
And over the 2,000km from Ashburn to Austin, that’s 3ms earlier. For some people, that’s really, really important. Humour them: they’re paying your bills.