Parallel Implementation of a FEM code by using MPI/PETSc and OpenMP Hybrid Programming Techniques
Abstract
The so called “hybrid parallelism paradigm”, that combines programming techniques for ar-
chitectures with distributed and shared memories using MPI (’Message Passing Interface’) and OpenMP (’Open Multi-Processing’) standards, is currently adopted to exploit the growing use of multicore computers in PC’s clusters, thus improving the efficiency of codes in such architectures (several multicore nodes or clustered symmetric multiprocessors (SMP) connected by a fast net to do exhaustive computations).
Meanwhile, for large-scale applications projects there is PETSc, a suite of data structures and routines for the scalable (parallel) solution of scientific applications modeled by partial differential equations, implementing MPI standard and widely used in parallel finite element codes development.
In this paper a parallel hybrid finite element code is developed and its performance evaluated, using MPI/PETSc for communication between nodes and OpenMP for parallelism within an SMP node. The cluster in which the code was run was the CIMEC’s ’Coyote’ cluster, which consists of 8 nodes with 8 cores each, connected through a Fast Ethernet net of 1Gb/sec.
chitectures with distributed and shared memories using MPI (’Message Passing Interface’) and OpenMP (’Open Multi-Processing’) standards, is currently adopted to exploit the growing use of multicore computers in PC’s clusters, thus improving the efficiency of codes in such architectures (several multicore nodes or clustered symmetric multiprocessors (SMP) connected by a fast net to do exhaustive computations).
Meanwhile, for large-scale applications projects there is PETSc, a suite of data structures and routines for the scalable (parallel) solution of scientific applications modeled by partial differential equations, implementing MPI standard and widely used in parallel finite element codes development.
In this paper a parallel hybrid finite element code is developed and its performance evaluated, using MPI/PETSc for communication between nodes and OpenMP for parallelism within an SMP node. The cluster in which the code was run was the CIMEC’s ’Coyote’ cluster, which consists of 8 nodes with 8 cores each, connected through a Fast Ethernet net of 1Gb/sec.
Full Text:
PDFAsociación Argentina de Mecánica Computacional
Güemes 3450
S3000GLN Santa Fe, Argentina
Phone: 54-342-4511594 / 4511595 Int. 1006
Fax: 54-342-4511169
E-mail: amca(at)santafe-conicet.gov.ar
ISSN 2591-3522