Climatologia de mesoescala em grade computacional / Mesoscale climatology in computacional grid

AUTOR(ES)
DATA DE PUBLICAÇÃO

2007

RESUMO

Establishing the climatology of meteorological models with ensemble techniques demands high computational power due to the huge amount of data to be processed. Being an application whose execution is independent, it is suitable for computational grid since it allows the computational effort to be shared with other geographically distributed machines. Computational grids consist of hardware and software infrastructure that provides trustworthy and consistent access to computational resources located at different geographical places. However, problems related to data transferring, storage and scheduling may influence application performance in a grid. G-BRAMS Project goals have been to develop a methodology for the generation of climatology in computational grid; to test 3 (three) grid platforms for this application: OurGrid, CIGRI/OAR and Globus; and to provide at least 10 years of the mesoscale model Regional Brazilian Atmospheric Modeling System (BRAMS) climatology for the community, for three regions of Brazil (North, Northeast and Southeastern/South). This grid comprises clusters installed at the Laboratório de Computação Aplicada (LAC), at the Centro de Previsão de Tempo e Estudos Climáticos (CPTEC) - belonging to the Instituto Nacional de Pesquisas Espaciais (INPE) - and at the Instituto de Informática (II) of the Universidade Federal do Rio Grande do Sul (UFRGS). However, although it has been proven that agglutinating resources in computational grid is viable and works for climatology generation using a portal, it was unknown if the application performance and the proposed use would be adequate for more extensive use. In this way, the objectives of the experiment described here have been: to implement the climatology developed by International Research Institute for Climate Prediction (IRI) in the computational grid; to understand how works the execution of the three-year monthly climatology (January-1996 to December-1998) of the BRAMS (three-year concept test) in the computational grid of G-BRAMS project; and to make a more refined analysis of the behavior and of the effective performance of the climatology in the environment of this grid, through the temporization of application processing in each node and data transmission. The results had shown that the developed scheduling works for a dedicated grid with dedicated portal. It was observed a performance fall of the application due to non-availability of the grid and to data communication time. It was verified that the methodology developed for the climatology is robust in a dedicated grid with dedicated portal. The question of data storage/transfer and of failures occurred on the grid must be better treated to guarantee the performance of the application.

ASSUNTO(S)

atmospheric models processamento de alto desempenho computação em grade grid computating meteorologia grid middleware infra-estrutura de grade meteorology high performance computing modelos atmosféricos computaÇÃo aplicada

ACESSO AO ARTIGO

Documentos Relacionados