摘要
在引进欧洲中期天气预报中心 (ECMWF)的全球谱模式的基础上 ,通过对原模式的分析改造 ,首次以分布与共享相结合的方式在国家气象中心IBM/SP高性能计算机上实现了全球谱模式的高效运行。采用调整向量长度、优化程序设计、完善消息传递机制和实现MPI与OpenMP的混合并行编程等方法 ,减少模式的通信量、计算量和内存的使用量 ,提高了计算效率。实现了在T2 1 3L31分辨率条件下 ,1 0天预报可以在 3h之内完成 ,达到业务对时限的要求。建立了与T2 1 3L31全球谱模式相配套的最优插值 (OI)并行处理分析系统 ,解决了由于观测站点在全球不均匀分布所带来的计算负载不均衡问题。在此基础上 ,实现了T2 1
Using combined method of distributed and shared memory parallelization, new global spectral model based on ECMWF is run on high performance computer IBM/SP with great efficiency in National Meteorological Center (NMC) for the first time. The model performance is improved greatly by means of changing vector length, optimizing program design, ameliorating message passing mechanism, realizing joint application of distributed and shared memory parallelization, to decrease communication, calculation and memory consumption. Now 10 day forecasts of the model with resolution T213L31 can be finished within 3 hours so as to satisfy operational time requirement. At the same time, the parallel Optimization Interpolation (OI) analysis system matching T213L31 global spectral model is constructed, and the problem of calculation load imbalance produced by spatial heterogeneousness of global observation stations is resolved, so that T213L31 global data assimilation and forecast system is built based on the above work, and so is corresponding automatic job control and watching system.
出处
《应用气象学报》
CSCD
北大核心
2004年第5期523-533,共11页
Journal of Applied Meteorological Science
