Latency Measurement to Hit Prime Time As Emphasis Shifts to 'How Fast'

The acceptance of high frequency trading techniques has shifted the emphasis around low-latency delivery of market data from a question of whether to implement high-speed connections to one of ‘how fast?’ Market practitioners expect to see greater activity around latency measurement, as clients seek proof that they’re getting the best performance out of their execution systems.

Some in the industry are calling for increased standardization of latency measurement, however, to allow market participants to compare the latency performance of their networks, execution systems and venue trading platforms on a like-for-like basis.

According to Henry Young, director of product development at latency measurement specialist TS-Associates, there are certain key steps that can be used to illustrate latency between specific points in the trade and/or market data delivery cycle. At the heart of the challenge, though, he says, is the issue of time synchronization, without which measuring time at milli-, micro- or nanosecond granularity is pointless.

Young says time synchronization issues start at the exchange or execution venue level. “Inaccurate and/or poorly defined time at exchanges makes it a headache for latency,” he says. “Exchanges publish time stamps, but they don’t say how they’re measuring it; it’s not a constant, there is no standard.” Venues are using a variety of different methods for time measurement, he says, from Network Time Protocol (NTP), Global Positioning Systems (GPS) and atomic clocks.

While millisecond timestamps are becoming standard among execution venues, Young says there remain problems with using that approach to tagging data.

First is the question of how clocks are synchronized. Exchanges make it difficult to understand which measure is being used by failing to make public the external time reference they use. Exchanges also need to make allowances for the possibility of their systems’ drifting out of sync even where they are using accepted external references, such as the UTC absolute time measure.

Second, there is no standard on where or when in the data delivery process the time stamp takes place, Young says. Is it at the point of order match, at the point the event data leaves the exchange servers, or at some other point in the execution process? Finally, the growing volumes of data are testing the usefulness of millisecond time-stamping. Young cites the peak Opra rate of 1,000 messages per millisecond.

Young believes that the situation could be improved if exchanges took steps to be more open about synchronization (by making public their base source of time), were specific about the event they’re time-stamping, and improve the granularity of their time-stamping.

TS-Associates has released Tradesync to address these issues. The system, announced during the summer and adopted by Chi-X Europe, uses Precise Time Protocol (PTP) IEEE standard, and is synchronized with GPS. The system allows execution venues to synchronize clients’ co-location servers to its time measure, while remote clients can synchronize using GPS. The result, Young reckons, is an accurate time-stamp with no performance degradation down to a granularity of 10 nanoseconds.

Key to this is Tradesync’s use of TS-Associates Application Tap, which uses a hardware card produced by Accolade to measure data events within the application server. Legacy latency measurement systems poll application systems for passing messages and then process at the file or server level, thereby adding latency. Application Tap, meanwhile, measures instances of memory access or writes and time-stamps these, with very little degradation of performance, Young says. This is particularly useful as firms incorporate more of the trading process into single, multi-core servers.

Application Tap monitors the time stamp of each event passing through the trading infrastructure up to a benchmark of 2 million events/second. It then pushes the data to TS-Associates’ TipOff latency management system for analysis.

Young believes that by modelling their processes more accurately, execution venues can make their trading environments more attractive to high-freqency trading clients. Increased granularity of data can give participants a better view of what’s truly happening in the marketplace, and this can attract players to a given venue, resulting in an increase in liquidity.