Desafios na paralelização de uma aplicação de câmara de mistura

Abstract

A program to perform simulations of a mixing chamber was modeled using the Navier-Stokes equations and discretized in a two-dimensional model. The program was implemented sequentially in Fortran90. Consequently it has an expressive computational time, taking hours to compute all the data. This work proposes the optimization of this program using parallel programming. To speed up code execution, parallel operations provided by the parallel programming interfaces OpenMP and OpenACC were used, using two meshes of different sizes as a case study. In this way, it was possible to calculate the performance gain when using APIs with different sizes of inputs for the problem, measuring how effective parallel programming has become for this problem, from a small mesh to a more complex, precise mesh and, consequently, more costly. With the test results, it was proved that the parallel version of the developed algorithm maintains the quality of the solution and reduces the execution time for the two meshes used in the experiments within the possible limits for the architecture and parallelizable parts of the code.