数值天气预报全球谱模式并行计算研究

Research on the Global Spectral Parallel Computing of Numerical Weather Forecasting

本文首先给出数值天气预报全球谱模式的控制方程;接着,利用数值天气预报全球谱模式计算的复杂性和数据整体相关性,给出了基于二维数据剖分方法的并行算法,并从负载平衡、减少通信延迟以及Cache与I/O优化三个方面给出了并行实现策略。

The paper first introduces the governing equations of the global spectral weather forecasting model, and then gives a parallel algorithm based on 2-D data partition by means of both the computational complexity of the global spectral model and the total correlation of spectral coefficients. Considering load balancing,communication latency reduction as well as cache and I/O optimization, we present a parallelization strategy.