TIM-2D fine-grain paralleling algorithms

Abstract

TIM-2D is a continuum mechanics code that uses irregular polygonal grids of an arbitrary structure. TIM-3D is paralleled at three levels. At the upper level, the simulation is paralleled in domains within the distributed memory model using MPI. At the second level, the simulation is paralleled in para-domains (the domain is broken down to fragments with cell arrangement into one layer), also using MPI. At the third (bottom) level, iterations of simulation cycles are paralleled within the distributed memory model using the OpenMP interface. These approaches can be used together in different combinations, or separately in a single simulation. The paper discusses fine-grain paralleling algorithms (the second level) and their interfacing with the domain decomposition and the OpenMP-based paralleling. The parallel efficiency is measured on the basis of a number of test simulations. The paper was prepared on the basis of the authors' report at the International Conference on Parallel Computing Technologies (PaVT-2009; http://agora.guru.ru/pavt).