Technological Aspects of the Hybrid Parallelization with OpenMP and MPI
- 964 Downloads
In this paper we present practical parallelization techniques for different explicit and implicit numerical algorithms. These algorithms are considered on the base of the analysis of characteristics of modern computer systems and the nature of modeled physical processes. Limits of applicability of methods and parallelization techniques are determined in terms of practical implementation. Finally, the unified parallelization approach for OpenMP and MPI for solving a CFD problem in a regular domain is presented and discussed.
KeywordsOpenMP Hybrid Parallelization Distributed Memory Computer System Twisted Factorization Unique Address Space
This work was supported by the Russian Foundation for Basic Research (projects 15-01-06363, 15-01-02012). The work was granted access to the HPC resources of Aix-Marseille Université financed by the project Equip@Meso (ANR-10-EQPX-29-01) of the program Investissements d’Avenir supervised by the Agence Nationale pour la Recherche (France).
- 4.Accary, G., Bessonov, O., Fougère, D., Gavrilov, K., Meradji, S., Morvan, D.: Efficient parallelization of the preconditioned conjugate gradient method. In: Malyshkin, V. (ed.) PaCT 2009. LNCS, vol. 5698, pp. 60–72. Springer, Heidelberg (2009). doi: 10.1007/978-3-642-03275-2_7 CrossRefGoogle Scholar
- 5.Saad, Y.: Iterative Methods for Sparse Linear Systems. PWS Publishing, Boston (2000)Google Scholar
- 7.Elizarova, T., Chetverushkin, B.: Implementation of multiprocessor transputer system for computer simulation of computational physics problems (in Russian). Math. Model. 4(11), 75–100 (1992)Google Scholar
- 8.Bessonov, O., Fedoseyev, A.: Parallelization of the preconditioned IDR solver for modern multicore computer systems. In: Application of Mathematics in Technical and Natural Sciences: 4th International Conference. AIP Conference Proceedings, vol. 1487, pp. 314–321 (2012)Google Scholar