Paralelização do algoritmo SUFI2: Uma abordagem Windows

Rodrigo de Queiroga Miranda, Josiclêda Domiciano Galvincio, Magna Soelma Besera de Moura, Raghavan Srinivasan,

Resumo


The Soil and Water Assessment Tool (SWAT) has been used for evaluating land use changes on water resources worldwide, and like many models, SWAT requires calibration. However, the execution time of these calibrations can be rather long, reducing the time available for proper analysis. This paper presents a Windows approach for calibrating SWAT using a multinodal cluster computer, composed of six computers with i7 processors (3.2 GHz; 12 cores), 8 GB RAM and 1 TB HDD each. The only requirement for this type of cluster is to have 64-bit processors. Our computers were setup with Windows Server HPC 2012 R2, a network switch 10/100, and regular Ethernet cables. We used the SUFI2 algorithm that comes with SWAT-CUP package to perform calibrations with 100 simulations at node level. Calibration runs were configured as follows: 1-12 (1 process interval), and 12-72 (12 processes interval), resulting in 17 runs. Each run was repeated three times, and results are presented as the mean execution time, in order to minimize any influence of resources fluctuations. Results showed that time of execution was reduced by almost half by using nine processes (15 min) in comparison with the one node control (28 min). We observed a linear decrease of execution time from one to nine processes. With additional processes, execution time increased about 23% and stabilized at 80% of the control. All processing is divided into five steps: distribute files (2.24% of all processing time), organize samples (0.89%), run SWAT (47.59%), collect results (46.51%) and cleanup (0.28%).


Palavras-chave


high performance; hydrology; modelling; water resources

Texto completo:

PDF (English)


DOI: https://doi.org/10.26848/rbgf.v10.5.p1535-1544



      

Revista Brasileira de Geografia Física - ISSN: 1984-2295

Creative Commons License
Esta obra está licenciada com uma Licença Creative Commons Attribution-NonCommercial 4.0 International License