Abstract: 1. Based on general parallel computing theory and the common characteristics of atmospheric models, an efficient scheme suitable for parallelizing serial numerical models is discussed in this paper, as well as the relevant problems that may arise. The proposed scheme includes four steps: model structure analysis, time cost analysis, data decomposition with communication design, and uniform debugging. These steps can help the numerical model programmer to complete the model's parallelization effectively. As an example, this scheme is used in the parallelization of the climate Spectral Atmosphere Model (SAMIL) with a resolution of R42L26 recently developed at LASG/IAP. According to the calculation structure of this model, different parallel skills are adopted in the Gaussian-grid and spectral spaces respectively. MPI is used to parallelize the physical grid-space calculation including all physical processes by dividing latitudes into several sections based on the model structure analysis. Moreover, some other processes such as time-integral and spectral-space calculations are also parallelized to improve the overall performance. Though this procedure is simple and can be implemented easily, the procedure's performance shows that the parallel efficiency is acceptable.2. Further, grid computing is also introduced in the article, and the method has been used for numerical climatic models in the real computing environment at IAP. In this paper, models are divided into two patterns, fine-grain and coarse-grain, according to the amount of data communication and are employed for the grid computing case studies. Experiments are carried out in several climatic models and quasi data-parallel models over both local and wide area networks. It is found that not all climatic models are suitable for grid computing: fine-grain models on a wide area network are not workable, but the efficiency of coarse-grain models is acceptable. A better computation effect can be achieved if the computing modes are organized more reasonably. These findings are most significant for climatic researchers in developing applications based on grid platforms in the next generation of supercomputing.3. Round-off error has an influence on the numerical model computations. It is easy to determine the differences between the results of the model SAMIL with different CPUs. After the analysis of the ten-year integration of SAMIL, the error range of global mean height at 500 hPa is determined; also, it is found that the global mean height error is decreased at the same level using double precision. The experiment described in the pape…

Key words: numerical model; parallel computing; parallelization scheme; grid computing; calculation order; multiple-precision computation; nonlinear dynamic system; roundoff error; maximum prediction time

# Parallel Computing for Climatic Numerical Models and Multiple Precision Computing for Nonlinear Dynamic Systems

This entry was posted in Master Thesis. Bookmark the permalink.