Implementação paralela para análises estáticas lineares pelo método dos elementos finitos

Abstract

In the finite element structural analysis, complex mechanisms representation in structures are increasing and requiring models with a great number of degrees of freedom. This situation represents a significant problem both in performance and memory usage on computers, particularly on phases such as assembling and solving of the equations system. Thus, there is a great need for high technology and performance computers. However, the use of these machines is very inefficient when considering its cost and, as the technology advances, the need for occasional replacement.

In order to solve this situation, attentions turned to the use of parallel computers. In this system modality, a big and complex problem is divided in smaller portions, that are individually solved in different computers. Therefore, parallel computing bring great advantages, such as smaller demand on components like memory and processor, besides the easiness to add new units on the system. Nonetheless, the concurrency demand the use of algorithms other than those from sequential computing, especially when high performance is needed. For the resulting time of a parallel computing be advantageous, the division of tasks must be performed in such way that all processes end at the same time along with minimum communication.

Considering the finite element method, procedures that perform a homogeneous division of elements among all processors and domains frontier minimization are essential. As a consequence of this, the requirement for well balanced workload and minimal communication is met.

This work addresses a finite element method parallel implementation. For this, the INSANE (Interactive Structural Analysis Environment) system, the Java programming language, the Object Oriented paradigm and the MPI (Message Passing Interface) pattern of data exchenge and communication are used.