The Preliminary Study on the Efficiency of Parallel Model Computation
-
摘要: 用有限区域中尺度差分网格模式,采用分区域并行计算方法,在TRANSPUTER并行计算机上研究大规模并行计算中计算问题复杂度与并行计算效率的关系。结果表明,采用适当的并行计算法,增加处理机个数和计算问题的复杂度可使并行加速比增大;并行效率则随着处理机个数的增加而降低,随着计算问题复杂度的增加而提高;大规模并行计算能明显地提高计算速度,很适合于解决大量计算的问题,在数值预报方面具有广阔的发展前景。Abstract: The suitability and relationship between complexity and efficiency of multiple parallel computation have been intensively studied by using the parallelized version of the limited-area fine-mesh grid-point model. The preliminary study indicates that, with suitable algorithm, the speedup can be improved by increasing the number of CPU (the number of transputers) in general. And, the efficiency of parallel may decrease with the CPUs, and increase with higher complexity. The results suggest that massively or highly parallel computer systems are most privileged for solving computationally intensive problem. In the light of its scalability the system which has the unlimited potential to provide much faster and more powerful computer resources required for further development of NWP and numerical modeling.
计量
- 摘要浏览量: 3693
- HTML全文浏览量: 218
- PDF下载量: 3355
- 被引次数: 0