Development of Basic Approach and Software Package for Effective Parallel Computing of Continuum Mechanics Problems on Hybrid Architecture Systems

Abstract

At present the requirements for calculation of continuum mechanics problems are very high: besides the complex geometry and various physical models, actual algorithms on large grids require high performance computing systems. Anyway, due to limited performance of “silicon” core, modern supercomputers represent set of cores and processors that work in parallel. Huge differences between sequential and parallel models lead to appearance of several parallelism stages: algorithmic, hardware and programming stages. It is necessary to reach demanding performance on each stage for development of effective program. Current parallel programming models provide several technologies; within the most popular ones are: OpenMP and MPI. Quite recently parallel technologies were replenished by General-purpose graphics processing units (GPGPU) technique. Each parallel programming model has its own special features. To attain high performance, parallel software should use several levels of parallelism within different models also taking into account its features. But it can be possible to build a programming technique when solving continuum mechanics problems.