Internet at the speed of light

The internet is such a slowpoke.

In principle, it should operate at nearly the speed of light, which is more than 670 million miles per hour. Instead, internet data moves 37 to 100 times slower than that. The technical term for this speed gap is “network latency,” the split-second delay in an internet connection as a signal travels from a computer to a server and back again.

We can do better, says Gregory Laughlin, a professor of astronomy in Yale’s Faculty of Arts and Sciences. Laughlin says we can make the internet at least 10 times faster — perhaps 100 times faster — in the United States.

Laughlin and colleagues P. Brighten Godfrey at the University of Illinois at Urbana-Champaign, Bruce Maggs at Duke, and Ankit Singla at ETH Zurich are co-leaders of an exploration into what is slowing the internet down — and what can be done to fix it. The project, funded by the National Science Foundation, is called Internet at the Speed of Light.

The researchers say a couple of key factors are holding the internet back. For example, the network of underground, fiber optic cable routes the internet depends upon is highly chaotic. It zig-zags beneath highways and railroad tracks, detours around difficult terrain such as mountains, and typically sends a signal hundreds of miles in the wrong direction at some point during a transmission.

Secondly, there’s the matter of the fiber optic cable itself, which is essentially glass. Internet data are pulses of light traveling through the cable; light moves significantly slower when it travels through glass.

Laughlin and his colleagues say a network of microwave radio transmission towers across the United States would allow internet signals to travel in a straight line, through the air, and speed up the internet.

Moreover, Laughlin says, this idea has already been successfully tested on a limited scale. For example, stock traders built a microwave network a decade ago between stock exchanges in Chicago and New Jersey in order to shave valuable microseconds off of high-frequency trading transactions.

In their final findings, which they presented at the 19^th USENIX Symposium on Networked Systems Design and Implementation  in April, Laughlin and his colleagues discovered that microwave networks are reliably faster than fiber networks — even in inclement weather — and that the economic value of microwave networks would make them worth their expense to build.

Laughlin spoke with Yale News recently about the project.

How did you come to be a part of Internet at the Speed of Light?

Gregory Laughlin: I was interested in the economic problem of where “price formation” in the U.S. financial markets occurs. This required the assembly and correlation of data from different markets, for instance the futures markets in the Chicago metro area and the stock markets in the New York metro area. When I began working on the problem [in 2008] it was clear that even when there was a strong motivation to cut latency down as much as possible between disparate locations, the physical telecommunications infrastructure still imposed limits that prevented signaling at speeds approaching the speed of light.

Why did this project appeal to you?

Laughlin: I like problems where physics, economics, and geography all intersect, and the problem of price formation is the perfect juxtaposition along those lines.

How is this approach different from other examinations of internet infrastructure?

Laughlin: A primary concern within studies of the physical structure of the internet is often bandwidth, where the concern is how much information per second one can transmit on a given line. Other work on latency has focused on ideas related to pre-positioning information, which is the idea behind content delivery networks. Our work takes the perspective of asking, “What would the solution look like if you wanted to speed up small-packet traffic as much as possible across the entire United States?”

What surprised you the most as you looked at what was slowing down the internet?

Laughlin: One thing, that’s very well known, but which never ceases to amaze me, is the enormous amount of information that can be carried on optical fibers. By transmitting light in different color bands simultaneously, single highly specialized multi-core glass fibers are now capable of carrying hundreds of terabits of data per second. My formative internet experiences occurred in the late 1980s and early 1990s, and so my current Yale office Wifi connection seems really fast. But it’s staggering to realize that a single fiber can now transmit data at a rate that exceeds my office connection by more than a factor of a million. It was thus surprising to realize that with the right hybrid infrastructure, the internet could be both extremely fast and capable of carrying staggering amounts of data. Yet because the internet has arisen in an organic way rather than a top-down pre-planned way, it turns out that there are all these curious pockets of slow performance.

You and your colleagues have suggested that a national network of microwave radio transmission towers would make the internet faster. Why is this?

Laughlin: Even though an overlay of microwave radio transmission towers would provide only a tiny, seemingly negligible increase in bandwidth for the U.S. internet, the overlay could handle an important fraction of the smallest, most latency-sensitive requests. This type of traffic is associated with procedures that establish a connection between two sites, and which involve a lot of back-and-forth transmissions that are a small number of bytes each. By speeding these up and taking the physically most direct routes, you can get a factor-of-10 to -100 increase for the traffic where it matters most. On the other hand, for applications like streaming video, where it’s possible to buffer the information, the microwave towers don’t need to be used. Fiber is the way to go if you have big blocks of data that need transferring.

What would it take, in terms of cost and commitment, to create such a network?

Laughlin: In our paper, we created a detailed model of a national microwave network that can transmit 100 gigabits per second between 120 U.S. cities at speeds that average just 5% slower than the speed of light [which provides the ultimate physical limit]. This network would involve roughly 3,000 microwave transmitting sites [that use existing towers], and we estimate that it would cost several hundred million dollars to construct.

Does that price tag make it worth doing?

Laughlin: We did a detailed cost analysis, and it seems very clear that a project of this type would provide an economic benefit. The applications run the gamut from things like telesurgery to e-commerce and gaming.

How often do you think about this as you download a document or click on a website?

Laughlin: Only when a site seems slow to load!

What reactions have you had to the project’s findings?

Laughlin: The team presented the findings at one of the leading conferences in the networking field, and the reaction was quite positive. Of course, it’s a big step from designing a network in theory and implementing it in practice. But we definitely feel that it’s something that would work and would be worth building.