高帧频面探测器分布式数据获取软件研究  

Research on distributed data acquisition software for high frame rate area detectors

作　　者：商琨琳 李正恒 鞠旭东 周悦 怀平 刘志 SHANG Kunlin;LI Zhengheng;JU Xudong;ZHOU Yue;HUAI Ping;LIU Zhi(ShanghaiTech University,Shanghai 201210,China;Shanghai Advanced Research Institute,Chinese Academy of Sciences,Shanghai 201210,China)

机构地区：[1]上海科技大学,上海201210 [2]中国科学院上海高等研究院,上海201210

出　　处：《核技术》2024年第10期106-114,共9页Nuclear Techniques

基　　金：国家自然科学基金委国家重大科研仪器研制项目(No.22127901)资助。

摘　　要：高帧频面探测器是上海硬X射线自由电子激光装置(Shanghai HIgh repetitioN rate xfel and Extreme light facility,SHINE)上主要成像类实验站的核心探测器,其预计数据通量将达到20 GB·s^(-1)以上。对于几十GB·s^(-1)原始数据流的实时接收和处理,传统单机系统很难应对。本文提出了一种多节点分布式面探测器数据获取和处理软件架构。首先,研究了不同网络库的性能,采用同步传输和CPU线程绑定等方法实现了接近3 GB·s^(-1)的单线程数据接收速率;依据X射线脉冲编号(Bunch ID)进行各模块数据的多节点分发与合并,采用4个节点实现了约23.5 GB·s^(-1)的并行事例重建速率;进一步,对数据刻度和压缩算法进行了实现和测试,结合bitshuffle和LZ4压缩算法,实现了约5.7倍的压缩率。本文验证了多节点分布式的并行数据获取方法的可行性,为后续面探测器高通量数据获取软件的开发打下基础。[Background]The high frame rate area detector is the core detector for the major imaging-based experimental stations at the Shanghai HIgh repetitioN rate xfel and Extreme light facility(SHINE),and its data throughput is expected to reach more than 20 GB·s^(-1).For the real-time receiving and processing of tens of GB·s^(-1) raw data,traditional single-machine systems are difficult to cope with.[Purpose]This study aims to propose a multi-node distributed data acquisition and processing software architecture for high frame rate area detector at imaging-based experimental stations of SHINE.[Methods]Firstly,the performance of different network libraries was investigated,and the synchronous transmission method combined with CPU thread binding was found to have the best single-thread data receiving performance.Then,a parallel event building method was introduced by simultaneously dispatching and combing different module data across multiple nodes based on Bunch ID.Furthermore,the data calibration and the bitshuffle/LZ4 compression algorithm were implemented and tested.[Results]Test results show that the highest single-thread data receiving rate is achieved at nearly 3 GB·s^(-1),a parallel event building data rate of approximately 23.5 GB·s^(-1) is achieved by using 4 server nodes,and the realized compression ratio is about 5.7.[Conclusions]The feasibility of the multi-node distributed parallel data acquisition method for high frame rate area detector is verified in this study,providing a foundation for the subsequent development of high-throughput data acquisition software for area detectors.

关 键 词：面探测器 数据获取 数据压缩 高帧频 分布式计算 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]