A comparison of PETSc library and HPF implementations of an archetypal PDS computation

M. Ethesham Hayder; David E. Keyes; Piyush Mehrotra

doi:10.1016/S0965-9978(97)00080-X

A comparison of PETSc library and HPF implementations of an archetypal PDS computation

M. Ethesham Hayder^*, David E. Keyes, Piyush Mehrotra

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

2 Scopus citations

Abstract

Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation - a nonlinear, structured-grid partial differential equation boundary value problem - using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSc, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPMD style for the PETSc library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introduction of concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.

Original language	English (US)
Pages (from-to)	415-423
Number of pages	9
Journal	Advances in Engineering Software
Volume	29
Issue number	3-6
DOIs	https://doi.org/10.1016/S0965-9978(97)00080-X
State	Published - 1998
Externally published	Yes

Bibliographical note

Funding Information:
This work was supported by the National Aeronautics and Space Administration under NASA Contract Nos NAS1-97046 and NAS1-19480, whilst the authors were in residence at ICASE MS 403, NASA Langley Research Center, Hampton, VA 23681-0001. Access to the NASA SP2s was provided under the NASA High Performance Computing and Communication Program.

Keywords

Nonlinear elliptic boundary value problems
Parallel languages
Parallel libraries
Parallel scientific computing

ASJC Scopus subject areas

Software
General Engineering

Access to Document

10.1016/S0965-9978(97)00080-X

Cite this

@article{95ee94c696aa47ddae1400bdbef392b8,

title = "A comparison of PETSc library and HPF implementations of an archetypal PDS computation",

abstract = "Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation - a nonlinear, structured-grid partial differential equation boundary value problem - using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSc, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPMD style for the PETSc library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introduction of concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.",

keywords = "Nonlinear elliptic boundary value problems, Parallel languages, Parallel libraries, Parallel scientific computing",

author = "Hayder, {M. Ethesham} and Keyes, {David E.} and Piyush Mehrotra",

note = "Funding Information: This work was supported by the National Aeronautics and Space Administration under NASA Contract Nos NAS1-97046 and NAS1-19480, whilst the authors were in residence at ICASE MS 403, NASA Langley Research Center, Hampton, VA 23681-0001. Access to the NASA SP2s was provided under the NASA High Performance Computing and Communication Program. ",

year = "1998",

doi = "10.1016/S0965-9978(97)00080-X",

language = "English (US)",

volume = "29",

pages = "415--423",

journal = "Advances in Engineering Software",

issn = "0965-9978",

publisher = "Elsevier Ltd",

number = "3-6",

}

TY - JOUR

T1 - A comparison of PETSc library and HPF implementations of an archetypal PDS computation

AU - Hayder, M. Ethesham

AU - Keyes, David E.

AU - Mehrotra, Piyush

N1 - Funding Information: This work was supported by the National Aeronautics and Space Administration under NASA Contract Nos NAS1-97046 and NAS1-19480, whilst the authors were in residence at ICASE MS 403, NASA Langley Research Center, Hampton, VA 23681-0001. Access to the NASA SP2s was provided under the NASA High Performance Computing and Communication Program.

PY - 1998

Y1 - 1998

N2 - Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation - a nonlinear, structured-grid partial differential equation boundary value problem - using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSc, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPMD style for the PETSc library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introduction of concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.

AB - Two paradigms for distributed-memory parallel computation that free the application programmer from the details of message passing are compared for an archetypal structured scientific computation - a nonlinear, structured-grid partial differential equation boundary value problem - using the same algorithm on the same hardware. Both paradigms, parallel libraries represented by Argonne's PETSc, and parallel languages represented by the Portland Group's HPF, are found to be easy to use for this problem class, and both are reasonably effective in exploiting concurrency after a short learning curve. The level of involvement required by the application programmer under either paradigm includes specification of the data partitioning (corresponding to a geometrically simple decomposition of the domain of the PDE). Programming in SPMD style for the PETSc library requires writing the routines that discretize the PDE and its Jacobian, managing subdomain-to-processor mappings (affine global-to-local index mappings), and interfacing to library solver routines. Programming for HPF requires a complete sequential implementation of the same algorithm, introduction of concurrency through subdomain blocking (an effort similar to the index mapping), and modest experimentation with rewriting loops to elucidate to the compiler the latent concurrency. Correctness and scalability are cross-validated on up to 32 nodes of an IBM SP2.

KW - Nonlinear elliptic boundary value problems

KW - Parallel languages

KW - Parallel libraries

KW - Parallel scientific computing

UR - http://www.scopus.com/inward/record.url?scp=0032041211&partnerID=8YFLogxK

U2 - 10.1016/S0965-9978(97)00080-X

DO - 10.1016/S0965-9978(97)00080-X

M3 - Article

AN - SCOPUS:0032041211

SN - 0965-9978

VL - 29

SP - 415

EP - 423

JO - Advances in Engineering Software

JF - Advances in Engineering Software

IS - 3-6

ER -

A comparison of PETSc library and HPF implementations of an archetypal PDS computation

Abstract

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this