@InProceedings{SouzaMend:2016:OpCoPe,
author = "Souza, Jefferson Gon{\c{c}}alves de and Mendes, Celso Luiz",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)}",
title = "Optimization of the computational performance of radiative
transfer model",
year = "2016",
organization = "Workshop de Computa{\c{c}}{\~a}o Aplicada, 16. (WORCAP)",
abstract = "In this work, ways are being sought to improve the computational
efficiency of the radiative transfer model BrasilSR. This model
was adapted to Brazilian conditions by LABREN/CCST/INPE, the main
feature of this group is to quantify the solar energy resource,
from incident solar radiation estimates from the Earths surface.
This is achieved by combining the approach of two streams in the
solution of a radiative transfer equation with the use of
climatological data and parameters determined from GOES family of
satellite images. The computer code presents results in models
used by the scientific community. However, due to its
computational cost, there is a high processing time involved when
such processing is performed in conventional processors (typically
a few days of processing). In the first phase of this study, an
analysis is being carried out of the performance of the original
version of the model - which is performed in sequential mode - in
order to indicate which processes are the most computationally
expensive ones. From these initial results, some parallelization
techniques are being explored in order to optimize their more
intensive sections, on the assumption that the parallelization
through the use of multiple threads in a shared memory environment
could bring improvements to the code. The implementation of this
kind of parallelization can be accomplished with standard OpenMP
directives, which allow the creation and automatic management of
threads, without much programming effort. The effectiveness of
this type of parallelization can be measured by model performance
comparisons as more threads are used in its execution.",
conference-location = "S{\~a}o Jos{\'e} dos Campos, SP",
conference-year = "25-26 out.",
language = "en",
targetfile = "souza_optimization.pdf",
urlaccessdate = "12 maio 2024"
}