基于OpenCL的超低延迟行情加速系统设计与实现  

DESIGN AND IMPLEMENTATION OF ULTRA-LOW LATENCY MARKET ACCELERATION SYSTEM BASED ON OPENCL

作　　者：冯一飞 丁楠 叶钧超 柴志雷 Feng Yifei;Ding Nan;Ye Junchao;Chai Zhilei(School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,Jiangsu,China;School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214122,Jiangsu,China;Jiangsu Provincial Engineering Laboratory of Pattern Recognition and Computational Intelligence,Wuxi 214122,Jiangsu,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122 [2]江南大学人工智能与计算机学院,江苏无锡214122 [3]江苏省模式识别与计算智能工程实验室,江苏无锡214122

出　　处：《计算机应用与软件》2025年第3期15-21,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61972180)。

摘　　要：针对量化高频交易中对数据处理的低延迟需求,定制一种超低延迟的行情系统,包含网络通信、数据解码和数据分析三部分功能,并在FPGA上进行实现。对各功能模块进行并行优化,构建全流水架构;对内存架构进行优化设计,提高数据传输速率;使用流接口实现模块间数据传输优化,降低数据处理延迟。实验结果表明:在Alveo U50上最大吞吐率可达38.4 Gbit/s,行情处理延迟最低为678 ns,波动稳定在10 ns间,与软件方案相比,性能提升12倍,吞吐率提升1.87倍,且延迟稳定。Aimed at the low-latency requirements for data processing in quantitative high-frequency trading,an ultra-low-latency market system is customized,which includes three functions of network communication,data decoding and data analysis,and is implemented on FPGA.Parallel optimization of each functional module was used to build a full pipeline architecture.The design of the memory architecture was optimized to increase the data transmission rate.The stream interface was used to optimize the data transmission between modules and reduce the data processing latency.Experimental results show that the maximum throughput rate on the Alveo U50 can reach 38.4 Gbit/s,the market processing delay is as low as 678 ns,and the fluctuation is stable within 10 ns.Compared with the software solution,the performance is improved by 12 times,the throughput rate is increased by 1.87 times,and the delay is stable.

关 键 词：VITIS OPENCL 现场可编程门阵列 高频量化交易 超低延迟 系统开发设计 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]