Keywords: solid mechanics, parallelization, iterative solver, fluid-structure interaction, MPI, OpenMP.

This paper introduces an hybrid parallel model of the solid mechanics simulation strategy of Alya-High Performance Computational Mechanics, a multi-physics code for supercomputing platforms, developed at Barcelona Supercomputing Center. The goal of the paper is the parallelization strategy, which is chosen as a hybrid MPI/OpenMP model in order to take advantage of all the levels of parallelism that a multicore architecture offers.

The paper describes the main features: numerical implementation, algorithms, solution schemes, parallel issues and code engineering. Space and time discretization are based on the finite element method and on finite differences respectively. The solution scheme presented is a total Lagrangian formulation for large deformations, with Newton schemes to cope with non-linearities. Time integration is done either explicitly or implicitly. To exploit the thread-level parallelism of multicore architectures, OpenMP parallelization is introduced in the most time-consuming routines of the solid mechanics, mainly the element assembly and the iterative solver. Additionally, as the communications between threads are via shared memory, there would be a reduction of the communication cost between processors. The main ideas implemented behind the parallel I/O aspects are detailed. Preliminary scalability results are presented for a three-dimensional beam test that requires large data structures. The results obtained are encouraging, outperforming the current MPI-based Alya implementation.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £40 +P&P)