摘要
江门中微子实验(JUNO)拥有当前世界上能量精度最高、规模最大的液体闪烁体探测器。缪子是JUNO的主要本底,每个缪子事例在大型探测器中产生百万量级的光子,但复杂的光子模拟计算量巨大,传统串行计算方式耗时较长。为此,提出一种基于GPU的分布式缪子快速模拟方法。利用多GPU卡并行加速闪烁光在液闪探测器中的传输过程,采用信息传递接口通信向多节点分发模拟任务和收集结果。测试结果表明,GPU方法具有良好的加速比,和CPU方法相比,加速比最高可达约250倍。
For the Jiangmen Underground Neutrino Observatory(JUNO)experiment,a liquid scintillator detector with the largest size and highest energy accuracy in the world is used.Cosmic Muon is one of the major backgrounds of the experiment and each of them generates millions of photons in the large detector.However,the complicated simulation of photons involves massive calculation and the traditional sequential computing is time-consuming.Thus,a fast Muon simulation method implemented on GPU is presented.Multiple parallel GPU cards are used to accelerate the transmission of scintillation photons in the liquid scintillator detector.Also,the Message Passing Interface(MPI)is used to distribute simulation tasks to nodes and collect processing results.The test results show that the GPU-based method has an excellent speedup ratio,which is up to about 250 times higher than that of the CPU-based method.
作者
易培淮
李卫东
林韬
邹佳恒
邓子艳
刘言
YI Peihuai;LI Weidong;LIN Tao;ZOU Jiaheng;DENG Ziyan;LIU Yan(School of Nuclear Science and Technology,University of Chinese Academy of Sciences,Beijing 100049,China;Institute of High Energy Physics,Chinese Academy of Sciences,Beijing 100049,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2021年第8期100-108,共9页
Computer Engineering
基金
国家自然科学基金青年科学基金(11805223)
中国科学院战略性先导科技专项(A类)(XDA10010900)
中国科学院青年创新促进会项目(2017021)。
关键词
GPU并行计算
分布式通信
信息传递接口
缪子模拟
江门中微子实验
GPU parallel computing
distributed communication
Message Passing Interface(MPI)
Muon simulation
Jiangmen Underground Neutrino Observatory(JUNO)
