|
Matematicheskoe Modelirovanie i Chislennye Metody, 2016, Issue 11, Pages 93–109
(Mi mmcm80)
|
|
|
|
Algorithm for computational performance improvement and processor load balancing to simulate the general atmosphere circulation
V. P. Parkhomenko Dorodnitsyn Computing Centre of the Russian Academy of Sciences, Moscow
Abstract:
The paper analyzes some factors affecting the parallel implementation performance of the atmospheric general circulation model designed on a cluster type multiprocessor computer. It considers several modifications of the initial parallel code of this model in order to improve both its computational efficiency and processor load balancing. The numerical scheme is modified according to the time of the atmospheric general circulation model for parallel computing of dynamics and physics blocks. The proposed procedure is used along with the procedures of paralleling the dynamics and physics blocks based on decomposition of the computational domain. It allows both optimizing the processor load balancing and increasing the paralleling efficiency. The data obtained while using the scheme for the physics block load balancing allow for complication of the physics block without increasing the total computational time. The results of numerical experiments are given.
Keywords:
Atmospheric general circulation model, parallel implementation features, numerical experiments.
Citation:
V. P. Parkhomenko, “Algorithm for computational performance improvement and processor load balancing to simulate the general atmosphere circulation”, Mat. Mod. Chisl. Met., 2016, no. 11, 93–109
Linking options:
https://www.mathnet.ru/eng/mmcm80 https://www.mathnet.ru/eng/mmcm/y2016/i11/p93
|
Statistics & downloads: |
Abstract page: | 87 | Full-text PDF : | 36 |
|