Inserir conteúdo.

In this work a parallel frontal solver is implemented and documented, based on an object oriented environment. The main objective of the project is to develop a solver which takes advantage of parallel architectures.

Four main classes are implemented, TPZFront responsible for the frontal matrix decomposition, TPZEqnArray stores a vector of sparse equations on TPZFrontMatrix, TPZEqnStorage that executes data storage and management (reading and writing to disk) and the fourth class TPZFrontMatrix that manages all three classes and the process as a whole.

Reading and writing to disk takes interface time (buffering), but little cpu time. Therefore if it is executed in a separate thread, the read/writing would occur in parallel with element matrix computation and front matrix decomposition. Element stiffness computation and front decomposition run in separate threads as well.

Performance results are shown to demonstrate the effectiveness of this approach on both Linux and Windows NT platforms. Shared memory parallelization is used.

A finite element program is described as a set of reusable components. Each component from which the finite element program is composed is described and its reusability-extensibility commented. Such an approach is felt to be ideal to foster scientific collaboration, because a researcher interested in using the object oriented programming philosophy for his/her research can use only this part of the environment which is of interest for his/her research.