High-Frequency Trading Calls for Precise Network Engineering
High-frequency trading firms are applying state-of-the-art network engineering in an effort to shave a few millionths of a second off their response times, thereby gaining a bead on competitors.
The methods vary in sophistication, from putting in fiber optic connections to more advanced techniques, such as operating system (OS) bypass.
“A wide range of technologies are being deployed,” said Bruce Tolley, vice president of solutions marketing at Solarflare, which provides hardware and software for low-latency applications.
“Some are rather technically simple, as in finding or building shorter optical paths between NY City and Chicago or between NY City and the City of London,” Tolley said.
For example, the latency of a link between a trading firm and an exchange can be reduced by co-locating with the exchange, or by building a passive optical link between the exchange and the trading firm’s server, making use of legacy metro “dark fiber” that was installed in the dot com boom of the late 1990s and early 2000.
“Firms are looking across the stack at microwave, FPGA, and continued enhancements in fiber paths/routes and network infrastructures,” said Emmanuel Doe, president of trading solutions at Interactive Data Corp.
OS bypass and direct memory access techniques can reduce latency to single digit micro seconds by giving the application direct access to the network.
“At the servers at the endpoints, the operating system itself contributes tens or more microseconds to the latency budget of a financial transaction,” said Tolley. “The deployment of OS bypass solutions can take out dozens of micros of latency.”
Niche service providers are building shorter paths between trading centers, while Solarflare and other high performance software vendor are delivering OS bypass solutions.
“For the big investments, like shorter paths between major trading centers, you might see consortia,” Tolley said.
Solarflare provides low latency software and hardware for high speed trading and precision time synchronization for application servers located in financial exchanges at key major trading centers (New York, London, Hong Kong, etc.)
OS bypass is a form of hardware acceleration by which an application can directly access input/output commands without invoking the operating system, enabling, the CPU to avoid having to switch between the application and OS, a process known as “context switching.”
It’s estimated that up to 40% of network overhead is attributable to context switching.
Solarflare’s OS bypass solution, OpenOnload, is an “accelerated network stack” that provides applications with direct access to network hardware.
“The network stack bypasses the operating system and significantly improves performance through the removal of context switches, which otherwise reduce the efficiency by which a processor can execute application code,” said Tolley.
This acceleration is most desirable for applications that are network intensive, such as market data and high-frequency trading applications.
Solarflare is also a leading vendor of “precision time synchronization” server adapters that enable trading customers to synchronize their networks locally, as well as across multiple international sites.
“You cannot increase the performance of system that you cannot measure,” Tolley said.
To measure, monitor and audit the performance of these networks in terms of transaction times and application latency, end user firms are deploying precision time protocol (PTP) with 10GbE networking, which enables the end user firm’s network to be synchronized with the exchanges within microsecond accuracy.
PTP replaces an older IP-based technology, network time protocol (NTP).
“With its millisecond precision, [PTP] enables for ultra-low latency and sub-microsecond clock synchronization,” said Tolley.
Algorithms have become more prevalent in the spot FX market.
QB’s Algo Suite for futures market trade execution is also being co-located to HKEX.
Breaking data silos is key to deploying automation beyond 'nuisance' orders.
They can be used on quantum hardware expected to be available in 5 to 10 years.
Streaming blocks change the basis of matching and price discovery so institutions can find new liquidity.