现场可编程门阵列
计算机科学
延迟(音频)
高频交易
实施
低延迟(资本市场)
嵌入式系统
软件
ARM体系结构
上市时间
高级合成
协议栈
调度(生产过程)
计算机体系结构
算法交易
操作系统
堆栈(抽象数据类型)
软件工程
计算机网络
电信
经济
运营管理
财务
作者
Andrew Boutros,Brett Grady,Mustafa Abbas,Paul Chow
标识
DOI:10.1109/reconfig.2017.8279781
摘要
High-Frequency Trading (HFT) systems require extremely low latency in response to market updates. This motivates the use of Field-Programmable Gate Arrays (FPGAs) to accelerate different system components such as the network stack, financial protocol parsing, order book handling and even custom trading algorithms. However, the long cycle of developing and verifying FPGA designs makes it challenging for HFT software developers to deploy such highly-dynamic systems, especially with their limited hardware design expertise. We present a complete highly-optimized infrastructure that implements low-latency system components in C++ using High-Level Synthesis (HLS). We also develop a framework that enables HFT algorithm developers to implement their trading algorithms in a high-level programming language and rapidly integrate it to the rest of the system. We implemented our HLS-based system on a Xilinx Kintex Ultrascale FPGA running at 156 MHz. Our on-board measurements show an end-to-end round-trip latency less than 870ns, which is comparable to that achieved by prior RTL-based implementations but requires reduced system development time and effort.
科研通智能强力驱动
Strongly Powered by AbleSci AI