PERFORMANCE OF HYBRID OpenMP/MPI PARALLEL PROGRAMMING APPLICATION OF FINITE ELEMENT METHOD

In the area of parallel processing and parallel algorithms several processors are used together to execute a single application faster. There are two major programming paradigms: Shared Memory and Message Passing. Each of them fits into a specific physical model, but there are multiprocessors architectures whose mapping to one of these paradigms is not so simple. SMP clusters, for example, are built connecting some shared memory machines through an interconnection network. Applications on SMP clusters can be programmed to use message passing between all processors. However, it’s possible to achieve better performance using a hybrid model which uses multithreading with shared memory communication inside the SMP node and message passing communication between nodes. The objective of the research presented in this paper was to implement and evaluate a hybrid model of parallel programming for an engineering application in order to evaluate and compare this model with a pure message passing version. Finite element formulation of linear elasticity problems results in linear equation systems that can be solved numerically by the conjugate gradient method. This article presents some numerical simulation results of a structural analysis, where the stiffness matrix of the system is a full matrix, using a 3D parallel code with two strategies for parallelization of the conjugate gradient method. The solution is obtained without any preconditioning technique to provide a performance comparison between a hybrid and a pure version of the same application.