Performance of a new CFD flow solver using a hybrid programming paradigm

Marsha Berger, M. J. Aftosmis, D. D. Marshall, S. M. Murman

Research output: Contribution to journalArticle

Abstract

This paper presents several algorithmic innovations and a hybrid programming style that lead to highly scalable performance using shared memory for a new computational fluid dynamics flow solver. This hybrid model is then converted to a strict message-passing implementation, and performance results for the two are compared. Results show that using this hybrid approach our OpenMP implementation is actually marginally faster than the MPI version, with parallel speedups of up to 599 out of 640 using OpenMP and 486 with MPI.

Original languageEnglish (US)
Pages (from-to)414-423
Number of pages10
JournalJournal of Parallel and Distributed Computing
Volume65
Issue number4
DOIs
StatePublished - Apr 2005

Fingerprint

OpenMP
Message passing
Computational fluid dynamics
Programming
Innovation
Paradigm
Data storage equipment
Hybrid Model
Hybrid Approach
Message Passing
Shared Memory
Computational Fluid Dynamics
Style

Keywords

  • Message passing
  • Parallel programming
  • Shared address space
  • Space-filling curves

ASJC Scopus subject areas

  • Computer Science Applications
  • Hardware and Architecture
  • Control and Systems Engineering

Cite this

Performance of a new CFD flow solver using a hybrid programming paradigm. / Berger, Marsha; Aftosmis, M. J.; Marshall, D. D.; Murman, S. M.

In: Journal of Parallel and Distributed Computing, Vol. 65, No. 4, 04.2005, p. 414-423.

Research output: Contribution to journalArticle

Berger, Marsha ; Aftosmis, M. J. ; Marshall, D. D. ; Murman, S. M. / Performance of a new CFD flow solver using a hybrid programming paradigm. In: Journal of Parallel and Distributed Computing. 2005 ; Vol. 65, No. 4. pp. 414-423.
@article{e443ade6e97d4696846d2bfd95d595db,
title = "Performance of a new CFD flow solver using a hybrid programming paradigm",
abstract = "This paper presents several algorithmic innovations and a hybrid programming style that lead to highly scalable performance using shared memory for a new computational fluid dynamics flow solver. This hybrid model is then converted to a strict message-passing implementation, and performance results for the two are compared. Results show that using this hybrid approach our OpenMP implementation is actually marginally faster than the MPI version, with parallel speedups of up to 599 out of 640 using OpenMP and 486 with MPI.",
keywords = "Message passing, Parallel programming, Shared address space, Space-filling curves",
author = "Marsha Berger and Aftosmis, {M. J.} and Marshall, {D. D.} and Murman, {S. M.}",
year = "2005",
month = "4",
doi = "10.1016/j.jpdc.2004.11.010",
language = "English (US)",
volume = "65",
pages = "414--423",
journal = "Journal of Parallel and Distributed Computing",
issn = "0743-7315",
publisher = "Academic Press Inc.",
number = "4",

}

TY - JOUR

T1 - Performance of a new CFD flow solver using a hybrid programming paradigm

AU - Berger, Marsha

AU - Aftosmis, M. J.

AU - Marshall, D. D.

AU - Murman, S. M.

PY - 2005/4

Y1 - 2005/4

N2 - This paper presents several algorithmic innovations and a hybrid programming style that lead to highly scalable performance using shared memory for a new computational fluid dynamics flow solver. This hybrid model is then converted to a strict message-passing implementation, and performance results for the two are compared. Results show that using this hybrid approach our OpenMP implementation is actually marginally faster than the MPI version, with parallel speedups of up to 599 out of 640 using OpenMP and 486 with MPI.

AB - This paper presents several algorithmic innovations and a hybrid programming style that lead to highly scalable performance using shared memory for a new computational fluid dynamics flow solver. This hybrid model is then converted to a strict message-passing implementation, and performance results for the two are compared. Results show that using this hybrid approach our OpenMP implementation is actually marginally faster than the MPI version, with parallel speedups of up to 599 out of 640 using OpenMP and 486 with MPI.

KW - Message passing

KW - Parallel programming

KW - Shared address space

KW - Space-filling curves

UR - http://www.scopus.com/inward/record.url?scp=14944351008&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=14944351008&partnerID=8YFLogxK

U2 - 10.1016/j.jpdc.2004.11.010

DO - 10.1016/j.jpdc.2004.11.010

M3 - Article

VL - 65

SP - 414

EP - 423

JO - Journal of Parallel and Distributed Computing

JF - Journal of Parallel and Distributed Computing

SN - 0743-7315

IS - 4

ER -