The Countdown to the Nanosecond

January 18, 2010
Alexa Jaworski

Does one millionth of a second—a microsecond— really matter? How about one billionth of a second? As exchanges and trading firms strive for ultra-low latency, a nanosecond most certainly can.

“Nanoseconds is the natural successor from microseconds, which was the natural successor from milliseconds,” said Donal Byrne, CEO of Corvil, a provider of electronic trading and market data latency management systems. “Likewise picoseconds will be the natural successor from nanoseconds. And so it is, in the world of low-latency and high-frequency trading.”

This month, BATS Exchange announced that it had decreased its average latency—the time it takes for the exchange to acknowledge or fill an order— to 250 microseconds.

“Two hundred fifty microseconds is a far cry from nanoseconds, but people would have looked at us five years ago and some of the major exchanges would have thought measuring in microseconds was probably crazy,” said Chris Isaacson, COO of BATS Exchange.

“Nanosecond granularity for measurement, I believe, has already become important for components within the trading lifecycle,” he said. “Each component will probably be measured in nanoseconds if not already, but the whole of an order’s lifecycle, I see being measured in microseconds for a ways out.”

DirectEdge does not have production metrics for its next-generation platforms yet, but upon launch, “we expect our performance will put us in the top tier of exchanges," said John Ryan, chief architect for the Jersey City, N.J.-based equity electronic communications network operator, which is currently in the process of becoming an exchange.

"Sub-microsecond (nanosecond) transaction latency is not attainable today and will likely not be feasible in the next 6-10 years,” said Ryan. “Beyond that is difficult to predict."

The NASDAQ Stock Market this month decreased the round-trip time for executing an order down to an average of 210 microseconds, and, on a light trading day, down to 190 microseconds, said Brian Hyndman, senior vice president of transaction services at Nasdaq OMX Group. “We’re down to the microseconds now, and that’s come upon us pretty quickly,” he said. “If you asked me a couple years ago if we’d be trading at 200 microseconds, I probably would have said no, so do I think we can go to nanoseconds in the next few years? I’d say possibly. If we stay on this current trend, we might get there as technology improves.”

Realistically, by the end of this year, the industry could see a lot of exchanges in the hundreds of microseconds, said Shawn Melamed, founder and CEO of latency measurement software provider Correlix. Nasdaq selected Correlix in November to provide Nasdaq OMX customers with independent real-time measurement of latency in trades on its market, for intraday and post-day analysis. “I think that’s the common goal, maybe someone would manage to break the 100-microsecond mark. Maybe we’ll have the first exchange going below 100, next year, maybe, it can get to tens of microseconds, but I think nanoseconds are a five-year horizon.”

Correlix measures up to one microsecond of granularity, which is pretty much what will be needed for the next two years, at least, said Melamed.

Competitor Corvil also supports nanosecond precision in its products, noted Byrne, who said many of his customers use this precision to examine nanosecond differences in arrival of messages. This is particularly important today in co-location facilities, he added.

“The final frontier is collocation, where the algorithms, the trading engines are right at the exchange, located as close as possible to where the orders are being executed so there’s almost no latency,” said Lloyd Altman, a senior executive in Accenture’s Capital Markets Industry division. “Whether that will be nanoseconds eventually, I would guess maybe one day it will be, but will it matter? What is the competitive advantage of being within nanoseconds? Ultimately, it really comes down to who got the order in first, who got the execution matched first, and the ultimate factor will be the speed of the execution engine within the exchange… you reach a point where it’s hard to see whether you are getting more efficient because it’s already so fast and so close.”

How far can information travel within a nanosecond?

The answer is 30 centimeters, according to Fergal Toomey, Corvil’s chief scientist. That’s 11.8 inches.

Therefore, noted Toomey, an information packet transmitted over electrical cable will travel one foot in one nanosecond. “In an electrical cable, the individual electrons themselves do not travel very fast, but the electrical signal does actually propagate very close to light speed,” he explained. In effect, it "hops" from electron to electron.  “Funnily enough,” added Toomey, “when optical fiber is used instead of electrical cable, the signal moves more slowly. This is because light travels more slowly in the glass or plastic used to make fibers than it does in a vacuum - the speed is reduced by about 1/3. So information in fiber will travel only 20 centimeters in one nanosecond.”

Nanoseconds do matter, but not for everyone, says Byrne. “Nanoseconds become relevant only if market centers can execute orders within nanosecond timeframes and you can achieve a relative latency advantage over competing traders,” he said. “No market centers are executing orders sub-microsecond, today. They will in the future and when they do, then nanosecond differences in accessing the matching engines of those markets will become important."

Still, those firms that can achieve nanoseconds temporarily have an edge—then someone else will come up with another edge, said Altman.

Once nanoseconds become the new benchmark, the race towards picoseconds—one-trillionth of a second— may not be light years away.