Record Setting Penguin Computing-based High-Frequency Trading Solution Wins HPC Wire Reader’s Choice Award for Best Use of HPC in Financial Services

November 2018 – Readers of the popular industry publication HPCwire voted a solution created with a Penguin Computing server running Solarflare®️, LDA Technologies®️, and Xilinx®️ technologies as their choice for best use of high-performance computing (HPC) in financial services for 2018.

Called the “Ultimate Trading Machine,” the solution set a new milestone for networking performance, achieving 98 nanosecond tick-to-trade latency, which is 18% faster than the previous record.

Announced at the SC18 conference in Dallas, Texas, the HPCwire Readers’ Choice Awards recognize the most outstanding individuals, organizations, products, and technologies in the industry.

Votes for a subscriber’s own organization are disqualified so, as Tom Tabor, chief executive officer (CEO) of Tabor Communications, publisher of HPCwire, said in a recent press release, “receiving an HPCwire Readers’ Choice Award signifies an undeniable amount of community support as well as continued success.”

Verified by the Securities Technology Analysis Center (STAC®️), the award-winning solution is a Penguin Computing®️ Relion®️ XE1112 1U, 19 inch server with Intel®️ Xeon®️ Scalable processors and optimized memory, bus, and storage options (including low-latency Intel Optane™ data center storage). Its interconnect is powered by the Solarflare XtremeScale 8722 software-defined network interface, delivering ultra-low latency and onload kernel bypass with offload to a Xilinx Kintex®️ UltraScale™ FPGA running the LDA Technologies Lightspeed TCP™ core software stack.

Per STAC, the key measurement is “actionable latency”, which is how long it takes the solution to start transmitting an order after receipt of a market tick, based on a model that simulates Chicago Mercantile Exchange trades. Lower latency results in the ability to execute individual trades more quickly and execute more trades in the same time period, a real competitive advantage.

Highlights of the results that make this possible include:

  • At 1Gbps, maximum Actionable Latency was just 98 nanoseconds for 68-byte frames and 109 nanoseconds for 507-byte frames.
  • At 1Gbps, maximum LIFO latency for 507-byte frames was -65 nanoseconds. (It was negative because the SUT was able to send simulated orders before the entire ingress packet was received.)
  • From the very low rate to the highest rate reported, across both message sizes, the maximum of each latency metric did not vary by more than 2 nanoseconds, which is within the measurement uncertainty of the test harness (+/- 2 ns).