Parallelization of the Structural Mechanics Solver b2000++pro: Assessment, Status and Future Strategy

Abstract

In this contribution we examine the parallel execution of the general-purpose Finite Element (FEM) structural mechanics solver b2000++pro. The solver is written in C++ and offers a wide range of different elements and models. It exploits thread parallelism wit Intel's Threading Building Blocks (TBB), and with OpenMP in the linear algebra solver. Distributed memory parallelism is achieved through the MPI parallelization provided by the linear algebra solver. This partial parallelization in MPI results in a split between the main, supervising process and all other processes that only act as workers for the linear algebra solver. We determine the optimal thread to MPI process ratio on compute nodes equipped with two AMD EPYC 7702 processors, each with 64 cores operating at 2,0 GHz. The current approach employs a single supervisor process that must hold the complete FEM model and perform all the computations outside of solving the linear algebra system. This poses an apparent obstacle to the scalability of the application on systems with distributed memory. We briefly discuss our approach to overcome this scalability obstacle and offer a partitioned computation of the complete program.