Otimização computacional do módulo de química do BRAMS através de técnicas de paralelismo em plataforma computacional multicore Xeon Scalable Processor

• Autor(a) principal: Borseti, Renato José Policani
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do LNCC
• Texto Completo: https://tede.lncc.br/handle/tede/288
• topic: Computação; Programação (Computadores); Processamento eletrônico de dados; Computação paralela; Modelo BRAMS; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Weather and climate prediction has a huge impact on human activities, from the safety of populations living in high-risk areas to severe weather events that cause serious health problems. In order to be able to carry out such forecasts, numerical time prediction models must be accurate and fast response. This work presents 6 strategies to optimize the computational performance of the regional model of BRAMS (Regional developments in the Regional Atmospheric Modeling System) in a machine with architecture of multicore of last generation through the module of this model. The chemistry of the atmosphere, fundamental to explain the modifications in the constitution of the gases and particles that compose the atmosphere, is governed by the partial differential equation of mass transport, whose discretization results in a system of ordinary differential equations of the type stiff because the concentration of the chemical species evolves with different orders of magnitude, being used the Rosenbrock method of 3a order in 4 stages for its resolution. At each stage of the method a system of linear equations must be solved, whose matrix corresponds to the chemical mechanism employed. This work has 3 contributions: i) The development and study of 6 strategies to reduce the execution time of the BRAMS model according to the architecture used, ii) the delivery of an optimized code, with a reduction of 58.87% in the runtime of the stages of the Rosenbrock method, which is already available on the CPTEC website in the version 5.4 and iii) instructions on how to use the strategies developed according to the machine’s processor architecture used.