计算机与现代化 ›› 2022, Vol. 0 ›› Issue (06): 49-55.

• 数据库与数据挖掘 • 上一篇    下一篇

气象高性能计算环境中模式协同研发管理

  

  1. (1.国家气象信息中心高性能计算室,北京100081;2.中国气象局地球系统数值预报中心业务运行室,北京100081)
  • 出版日期:2022-06-23 发布日期:2022-06-23
  • 作者简介:赵春燕(1984—),女,北京大兴人,高级工程师,CCF会员(H3172M),硕士,研究方向:气象高性能计算应用支撑软件设计,E-mail: zhaocy@cma.gov.cn; 通信作者:孙婧(1971—),女,研究员,硕士,研究方向:气象高性能计算系统设计与资源调度管理,E-mail: sunj@cma.gov.cn; 通信作者:胡江凯(1973—),男,研究员,硕士,研究方向:数值天气预报系统开发,E-mail: hujk@cma.gov.cn; 周斌(1970—),男,高级工程师,硕士,研究方向:天气数值预报业务系统设计开发,E-mail: zhoubin@cma.gov.cn。
  • 基金资助:
    国家重点研发计划项目(2017YFA0604500,2017YFC1501903)

R&D Collaborative Management of Meteorological Model in High Performance Computing

  1. (1. Dept. of High Performance Computing, National Meteorological Information Centre, Beijing 100081, China;
    2. Dept. of Business Operation, CMA Earth System Modeling and Prediction Centre, Beijing 100081, China)
  • Online:2022-06-23 Published:2022-06-23

摘要: 面向气象科学的数值预报模式的研发过程是一个多学科交叉、持续改进型的复杂系统工程。在地球系统模式、E级计算和后摩尔时代的发展趋势下,气象数值模式研发协同面临更复杂的协同、更专业的计算平台调试分析、更广泛的共享应用等挑战。从以上需求和挑战出发,在气象高性能计算环境中,采用Git分布式技术、Python及工作流技术,建设气象数值模式研发协同管理支撑环境和标准,实现模式协同研发过程的管理、成果的集成共享和研发调试试验及分析一体化的支撑,以提升模式研发协同效率和业务化效率,保障研发成果的完整性和可跟踪性,提升大规模科学软件研发的管理能力。应用效果表明,本研究规范化了模式研发的协同过程,并建立流畅的协同支撑环境,提升模式迭代升级和业务化效率,能够为科学研究、科学计算等大型传统科学计算模型的持续研发改进管理及软件工程协同管理提供借鉴。

关键词: 气象高性能计算, 研发管理, 协同, 数值模式, 派-曙光, 中试

Abstract: The process of numerical prediction model research and development is a multi-disciplinary complex system engineering. As Moore’s law approaches its limit , exascale computing is coming and with the trend of earth system, the R & D collaboration of meteorological numerical model faces more complex collaboration, more professional computing platform to debug, wider shared demand challenges. In view of the above challenges, application research has been carried out around the solution of collaborative management, R & D process, code integration and sharing, model debug and experiment test-bed in high-performance computing environment with Git, Python and workflow to improve the collaborative efficiency and operational efficiency. The results show that this research standardizes the R & D process and achievement management, supports the convenient R & D management, debugging and analysis, improves the efficiency of upgrading and operation, it can provide reference for large-scale traditional scientific research on high performance computing.

Key words: meteorological HPC, SCM, collaboration, NWP model, PI-Sugon, test-bed