An evaluation of MPI and OpenMP paradigms in finite‐difference explicit methods for PDEs on shared‐memory multi‐ and manycore systems

Abstract

Summary This paper focuses on parallel implementations of three two‐dimensional explicit numerical methods on Intel® Xeon® Scalable Processor and the coprocessor Knights Landing. In this study, the performance of a hybrid parallel programming with message passing interface (MPI) and Open Multi‐Processing (OpenMP) and a pure MPI implementation used with two thread binding policies is compared with an improved OpenMP‐based implementation in three explicit finite‐difference methods for solving partial differential equations on shared‐memory multicore and manycore systems. Specifically, the improved OpenMP‐based version is a strategy that synchronizes adjacent threads and eliminates the implicit barriers of a naïve OpenMP‐based implementation. The experiments show that the most suitable approach depends on several characteristics related to the nonuniform memory access (NUMA) effect and load balancing, such as the size of the MPI domain and the number of synchronization points used in the parallel implementation. In algorithms that use four and five synchronization points, hybrid MPI/OpenMP approaches yielded better speedups than the other versions did in runs performed on both systems. The pure MPI‐based strategy, however, achieved better results than the other proposed approaches did in the method that employs only one synchronization point.

An OpenMP‐based breadth‐first search implementation using the bag data structure

“…17 We also plan to run the implementation on machines with the third and fourth generations of Intel ® Xeon ® Scalable processors. We also intend to use the MPI standard for inter-socket communication and the OpenMP standard for intra-socket communication, as performed by Cabral et al 19 We also plan to run these implementations using several nodes of the SDumont supercomputer. 20…”

Implementação de overthread em uma aplicação de método numérico HOPMOC através de um algoritmo híbrido MPI/OpenMP

“…Para verificar se estratégia OMP-EWS apresentaria as mesmas vantagens em outros algoritmos similares, Cabral et al [Cabral et al 2019] compararam três métodos numéricos bidimensionais em diferentes processadores Intel R . Nesse trabalho, foram utilizadas versões híbridas MPI e OMP-EWS dos métodos Hopmoc, Diferenças Finitas Totalmente Explícita para as equações do Calor e de Laplace bidimensionais.…”

“…No algoritmo 1, mostra-se uma visão geral do pseudocódigo do método Hopmoc. A versão híbrida do método Hopmocé uma estratégia que faz uso tanto de threads OpenMP quanto de processos MPI para dividir a carga de trabalho e permitir uma paralelização eficiente [Cabral et al 2019]. A malha de entradaé inicialmente dividida em seções menores, de forma igualitária.…”

“…A implementação do método Hopmoc bidimensional por meio da estratégia MPI/OMP-EWSé baseada no algoritmo implementado em uma publicação recente [Cabral et al 2019]. Essa abordagem conta com um modelo de paralelização constituído de dois mecanismos principais: a divisão explícita de trabalho (EWS) e a sincronização entre threads adjacentes (AdjSync).…”

High‐performance computing for computational science

“…Partial differential equations (PDEs) are widely employed in many scientific and engineering applications. Cabral et al 4 studied the performance of three numerical methods for solving PDEs on a 2D domain, and evaluated their performance on two shared memory architectures, namely, the multi‐core SKL and the manycore KNL. The authors studied several implementations combining OpenMP and MPI, and the best configuration depends on characteristics such as the program size, the number of synchronization points, and characteristics related to the NUMA architecture.…”