“The Ultimate Trading Machine” from Penguin Computing sets Record for Low Latency

Print Friendly, PDF & Email

The world of High Frequency Trading is all about reducing latency to make money. At the recent STAC Summit in Chicago, a Penguin Computing device called The Ultimate Trading Machine achieved a record-low 98 nanosecond tick-to-trade latency, some 18% faster than the previous world record.

“The time interval between receiving a market tick and sending a buy or sell order is the tick-to-trade latency. Decreasing the tick-to-trade latency improves the queue positions of trades. This results in higher probabilities for trades to be executed, and, thus, delivers a competitive advantage.”

The Ultimate Trading Machine was evaluated using the STAC-T0 benchmark suite. This benchmark simulates the time for the Chicago Mercantile Exchange to emit a trade data packet and for the Exchange to receive a market order packet. This is the minimum, possible trade execution time. STAC-T0 is designed to isolate the tick-to-trade network I/O latency and to measure this latency with high precision (an uncertainty of just ±2 nanoseconds is incorporated into reported results).

The solution achieves ultra low processing latencies through optimal design choices and select components. It is built on a solid performer, the Penguin Computing Relion XE1112 server. This 1U, 19″ EIA form-factor server features dual Intel Xeon Scalable processors and optimized memory, bus, and storage options (including low-latency, Intel Optane data center storage). The 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.

Facing tremendous pressures to optimize the transaction lifecycle, the financial services industry helps drive innovations in many core technologies. At Penguin Computing, we empower our customers with with open technology solutions that achieve performance requirements while keeping costs low and avoiding vendor lock-in. We are proud to join with our partners to deliver this Ultimate Trading Machine.

The 98 nanosecond result was achieved at 1 gigabit per second ingress for 68-byte frames and 109 nanoseconds for 507-byte frames. This is the maximum actionable latency, or how long it takes the solution to start transmitting an order after it has received all the necessary information to evaluate whether to send an order. Under repeated tests, the maximum latency metric remained consistent, varying no more than 2 nanoseconds, or within the uncertainty of the test harness itself. These results demonstrate the solid, record-setting performance of the Ultimate Trading Machine.

Download the STAC-T0 Report

Sign up for our insideHPC Newsletter