A high scalability parallel algebraic multigrid solver

Abstract

In this paper we present the performance analysis of a parallel Algebraic Multigrid Solver (AMG) for a finite volume unstructured CFD code. The multigrid solver is part of an unstructured cell-centered finite volume code. The parallelization of the solver is based on the domain decomposition approach using the single program multiple data paradigm.The Message passing interface Library (MPI) is used for communication of data. An ILU(0) iterative solver is used for smoothing the errors arising within each partition at the different grid levels, and a multi-level synchronization across the computational domain partitions is enforced in order to improve the performance of the parallelized Multigrid solver. A number of Multigrid cycles (V, W and F-Cycle) and two strategies have been implemented. Tests on different grids have been conductied for a non-linear problems on up to 28 parallel processors. Results show, that synchronization plays across partitions for multigrid-levels plays an essential role in ensuring good scalability. Furthermore for large partition number gathering coefficients across partitions is important to ensure a convergence history on par with the sequential solver.