您当前位置:首页 >> 研究内容

研究内容

研究内容
Research Content
一、数值并行算法及性能优化技术研究
以提高模式计算效率为目标,研究地球系统模式的高效数值并行算法,结合非数值并行算法的成果,研制高效数值软件模块,进行性能优化,形成支撑地球系统模式高分辨率、上千核计算的高效数值并行算法库。
1) Research on Numerical Parallel Algorithms and Performance Optimization Technologies
To improve the computational efficiency and ability of the Earth System Model, we will develop highperformance,numerical, parallel algorithms including adaptive mesh refinement (AMR), pre-conditioner,high-dimensional Fast Fourier Transform (FFT), and high-performance solvers of certain special partial
differential equations (PDEs). We will do the research on new parallel techniques, including GPU and CELL,and we will implement performance optimization methods such as locality of data access, vectorization, and instruction-level optimization. Finally, we will build high-performance numerical algorithm libraries to be used for high-resolution, thousand-core modeling of the ESM.

 

二、地球系统模式并行耦合技术研究
面向地球系统模式多个分量的耦合,研制高效的物理量插值并行算法、通量并行算法和子区域映射并行算法,建立通用耦合并行算法库,为模块化并行耦合器的研制提供算法支持。
2) Research on parallel coupling technologies for the ESM To efficiently couple multiple components of ESM, this subproject will develop parallel coupling technologies , including parallel remapping algorithms, parallel flux algorithms, parallel decomposition algorithms, and the generic library for these
algorithms.

 

三、地球系统模式并行应用框架
面向地球系统模式各分量采用的结构网格,结合局部网格自适应嵌套,研究非数值并行算法,通过继承、集成和发展高效软件模块,研制适应上千核的地球系统模式并行应用框架JEarth和模块化二维并行耦合器。
3) Parallel Applications Infrastructure for the ESM
Focusing on the Earth System Model based on structured meshes, a parallel application infrastructure called J-Earth and a modularized two-dimensional parallel coupler, which runs efficiently on supercomputers with thousands of cores, will be developed. In particular, useful adaptive local refinement techniques and efficient nonnumerical parallel algorithms will be researched, and high-quality software modules will be encoded by means of inheritance, integration and development.

 

四、高效物理气候系统并行模式的研发
实现大气和海洋模式在J-Earth上的移植和发展,采用新模块化二维并行耦合器,发展可以使用上千核的高效物理气候系统并行模式H-FGOALS,并建立其初始化系统,为年代际气候预测提供有效的工具。

4) Development of a high-performance physical climate system parallel model
Based on the reconstruction of the atmospheric and oceanic models in J-Earth and the two-dimensional modular parallel coupler developed by subproject 2, a high-performance physical climate system model (CSM) H-FGOALS and its initialization system will be created, the latter being an effective tool for decadal climate prediction.

 

五、物理气候系统模式的典型应用示范

利用H-FGOALS和已有物理气候系统模式,依托国内百万亿次高性能计算机,开展年代际和百年际气候变化预测与预估试验(CMIP5)。通过分析对比,评估新建模式的并行效率、可扩展性以及并行算法可靠性。

5) Typical application demonstration of the physical climate systemmodel

Using the H-FGOALS model and the current physical CSMs, and relying on the domestic hundreds-of-TFLOPS high-performance computers, the Coupled Model  Intercomparison Project Phase 5 (CMIP5) experiments for decadal climate prediction and multi-century climate projection will be conducted; the parallel computing efficiency, scalability and reliability of parallel algorithms for the newmodel will be analyzed and assessed.

通讯地址:北京市朝阳区华严里40号 北京9804信箱 中国科学院大气物理研究所科研楼
电话:010-82995223 010-82995173