Paralelização de algoritmo de simulação de Monte Carlo para a adsorção em superfícies heterogêneas bidimensionais

• Autor(a) principal: Gomes, José Luiz de Souza
• Data de Publicação: 2009
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Estadual de Maringá (RI-UEM)
• Texto Completo: http://repositorio.uem.br:8080/jspui/handle/1/2543
• topic: MPI; Processamento paralelo; Computação científica; Paralelização; Particionamento de domínio; Algoritmo de Monte Carlo; Adsorção; Simulação; Clusters; Computação.; MPI; Parallel processing; Scientific computing; Monte Carlo method; Simulation; Adsorption.; Ciências Exatas e da Terra; Ciência da Computação

This work discusses issues related to the parallelization of sequential algorithms used in the solution of scientific problems, many of them written in C or FORTRAN, in a time where the parallel and distributed programming facilities, in hardware and in software, were not as available as nowadays. Many of these algorithms require a long execution time, even though they present important results during simulations. Areas such as Physics, Biology and Engineering can benefit from parallel and distributed execution of these algorithms on computer clusters, which can be acquired at low cost. Consequently, a larger volume of data can be processed, providing new results and enabling the research progress in scientific areas. The objective of this work is to parallelize a Chemical Engineering scientific algorithm for molecular adsorption on two-dimensional heterogeneous surfaces. This algorithm uses the Monte Carlo method to calculate the energy state of the system after molecular movements and the results are used to draw isotherm diagrams, comparing them with real experiments and known data. Therefore, questions about the task of parallelization found in the literature were studied and implemented based on the model suggested by Foster (1995). Four parallel versions have been implemented and discussed the different approaches taken in each one, such as domain partitioning, dynamic load allocation and fault-tolerance. Its sequential execution spends long processing time and the parallel versions showed a reduction of execution time by approximately 73.7%, 73.4%, 80% and 83.17%, respectively, where the 4th version is the most efficient, making better use of the available resources in the parallel environment. Thus, simulations with larger volume of data could be made. It is expected, therefore, that the results will be more significant for the area to which it applies.