Современные информационные технологии и IT-образование (Oct 2022)

Comparative Performance Analysis of MPI- and OpenMP-Programs on the Example of Parallel Calculations in the Framework of the Nucleus-Nucleus Potential Model and the φ0-Spintronic Model

  • Maxim Bashashin,
  • Elena Zemlyanaya

DOI
https://doi.org/10.25559/SITITO.18.202203.545-557
Journal volume & issue
Vol. 18, no. 3
pp. 545 – 557

Abstract

Read online

The results of a study of the performance of problem-oriented programs developed using the MPI and OpenMP parallel programming technologies which implement the numerical solution of problems in the framework of two models that are actively used in nuclear physics and the physics of superconducting structures are presented. The first task is the construction of the nucleus-nucleus scattering potential based on the double folding model (MDF), which is reduced to the numerical solution of the non-linear integral equation with integrals of multiplicity 4 by the fix point method. The second task is the simulation of the magnetic moment reversal in the Josephson φ0-junctions of the system "superconductor – ferromagnet – superconductor" (SFS) on the plane of model parameters. The calculation here is reduced to a multiple numerical solution of the Cauchy problem for the corresponding system of ordinary differential equations for the values of the parameters running over the specified intervals with some step. Despite the cardinal differences in the goals and methods of numerical research, both problems have common properties – relatively easy to implement parallelism and not too high requirements for computer resources. For both tasks, the results of comparative calculations of MPI and OpenMP versions on different computing systems using different compilers and different numbers of MPI processes and OpenMP threads are presented. It is shown that for problems of this type, both technologies generally provide comparable characteristics in terms of a speedup of calculations and the minimum achievable computation time with an increase in the number of involved parallel MPI processes and OpenMP threads. The calculations were carried out using the computing resources of the Multifunctional Information and Computing Center of the Mescheryakov Laboratory of Information Technologies of the Joint Institute for Nuclear Research.

Keywords