MPI在蒙特卡罗程序GMT中的应用和发展

发布时间：2018-08-05 12:05

【摘要】：针对ADS颗粒靶概念的研究和设计,中国科学院近代物理研究所自主研发了蒙特卡罗模拟软件GMT。为了提高GMT程序的计算效率,研究了MPI在GMT中的应用和发展,实现了大规模随机数在进程中的随机分配,并采用快速读写文件的方式替代了MPI相关数据通信函数,极大地提高了计算效率。并研究了不同规模计算实例进程数、加速比、效率之间的关系,确定了最大加速进程数及并行效率最高时的进程数,为科研工作者在计算资源和计算效率之间选择最优计算方案提供了科学依据。MPI在GMT中的成功应用使计算资源得到了充分、高效的利用,极大地提高了计算效率,解决了蒙特卡罗方法中大规模事件模拟计算时间长、计算不稳定等问题,在散裂靶大规模扫描计算中发挥了重要的作用。
[Abstract]:Aiming at the research and design of the concept of ADS particle target, the Institute of Modern Physics of the Chinese Academy of Sciences has independently developed the Monte Carlo simulation software. In order to improve the computational efficiency of GMT program, the application and development of MPI in GMT are studied, the random allocation of large scale random numbers in the process is realized, and the method of fast reading and writing files is used to replace the MPI related data communication function. The calculation efficiency is greatly improved. The relationships among the number of processes, the speedup ratio and the efficiency are studied, and the maximum number of accelerated processes and the number of processes with the highest parallel efficiency are determined. This paper provides a scientific basis for scientific researchers to choose the optimal computing scheme between computing resources and computing efficiency. The successful application of MPI in GMT makes computing resources fully and efficiently utilized, and greatly improves the computational efficiency. It solves the problems of long time and unstable calculation of large-scale event simulation in Monte Carlo method and plays an important role in the large-scale scanning calculation of spallation target.
【作者单位】：中国科学技术大学;中国科学院近代物理研究所;
【基金】：中国科学院战略性先导科技专项(XDA03030100)~~
【分类号】：TL329.2

【参考文献】