Advanced Computing in the Age of AI | Friday, March 29, 2024

Servers On Steroids Speed Stock Swaps 

(Vintage Tone/Shutterstock)

Executing stock trades often comes down to being first in the queue of traders. Hence, developers of high-frequency trading platforms that connect trading servers to networks focus on reducing latency to nanoseconds to boost so-called "tick-to-trade" networking performance.

Solarflare, which specializes in connecting trading servers to Ethernet networks, claimed this week it has along with FPGA and high-end computing partners managed to reduce electronic trade latency to 98 nanoseconds. Those results are based on industry benchmark testing that measures networking performance.

Solarflare, Irvine, Calif., said it worked with FPGA specialist Xilinx along with FPGA product developer LDA Technologies and HPC vendor Penguin Computing. Together, they built a server configuration called the "Ultimate Trading Machine" that underwent testing using a new benchmark proposed by the Securities Technology Analysis Center. The benchmark, STAC-TO, is described as a measure of how fast a trading platform can move critical data into and out of an application on an Ethernet network.

According to the testing center, the trading platform developed by Solarflare and its partners received 68-byte Ethernet frames at 1 gigabit per second. The server configuration was able to transmit a trading order with a "maximum actionable latency" of 98 nanoseconds. STAC said the benchmark test had a margin of error of +/- 2 nanoseconds.

Penguin Computing's server used in the benchmark testing included dual Intel Skylake processors, Xilinx Kintex FPGAs running on LDA Technologies' FPGA-based IP core. The platform was linked with Solarflare's software-defined network interface controller.

Earlier this year, Solarflare said it had reduced network connectivity portion of the tick-to-trade life cycle to about 22 nanoseconds. The second portion of the trading life cycle is making the trade when a stock tick, or price, comes up on the trading wire. High-frequency trading systems use sophisticated algorithms, and Solarflare employed Xilinx FPGA's to speed the algorithms.

"The time interval between receiving a market tick showing an opportunity to an algorithm, and sending the buy/sell order is total tick-to-trade latency," explained Ahmet Houssein, Solarflare's vice president of marketing. "The latency of network input-output is a crucial factor in total tick-to-trade latency."

The goal for brokers is getting clients to the front of trading queues in fewer nanoseconds than competitors. "Everyone wants to be at the top of the queue in a first-in, first-out matching engine," noted Rob D’Arco, CEO of Rival Systems, which has been using Solarflare's network controllers since 2015.

Penguin Computing noted that the arrival of servers based on scalable Intel Xeon processors was a key component for building the faster trading platform, which Solarflare said Tuesday (Oct. 24) is available now.

Solarflare's server connectivity technology is designed for neural networks running sophisticated financial trading algorithms. The company claims its framework is used by major global stock exchanges, commercial banks and hedge funds.

 

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).

EnterpriseAI