Performance of the 3D FFT on the 6D network torus QCDOC parallel supercomputer

Bin Fang; Yuefan Deng; Glenn Martyna

doi:10.1016/j.cpc.2006.12.006

Performance of the 3D FFT on the 6D network torus QCDOC parallel supercomputer

Bin Fang, Yuefan Deng, Glenn Martyna

Mathematics

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

Abstract

QCDOC is a massively parallel supercomputer with tens of thousands of nodes distributed on a six-dimensional torus network. The 6D structure of the network provides the needed communication resources for many communication-intensive applications. In this paper, we present a parallel algorithm for three-dimensional Fast Fourier Transform and its implementation for a 4096-node QCDOC prototype. Two techniques have been used to increase its parallel performance: simultaneous multi-dimensional communication and communication-and-computation overlapping. Benchmarking experiments suggest that 3D FFTs of size 128 × 128 × 128 can scale well on such platforms up to 4096 nodes. Our performance results suggest stronger scalability on QCDOC than on IBM BlueGene/L supercomputer.

Original language	English (US)
Pages (from-to)	531-538
Number of pages	8
Journal	Computer Physics Communications
Volume	176
Issue number	8
DOIs	https://doi.org/10.1016/j.cpc.2006.12.006
State	Published - Apr 15 2007

Keywords

Ab initio calculation
Blue Matter
FFT
Long-range interaction
Molecular dynamics simulation
Parallel efficiency
Strong scalability

ASJC Scopus subject areas

Hardware and Architecture
General Physics and Astronomy

Access to Document

10.1016/j.cpc.2006.12.006

Cite this

@article{af0013a75f4649aa9ab6420176ef1387,

title = "Performance of the 3D FFT on the 6D network torus QCDOC parallel supercomputer",

abstract = "QCDOC is a massively parallel supercomputer with tens of thousands of nodes distributed on a six-dimensional torus network. The 6D structure of the network provides the needed communication resources for many communication-intensive applications. In this paper, we present a parallel algorithm for three-dimensional Fast Fourier Transform and its implementation for a 4096-node QCDOC prototype. Two techniques have been used to increase its parallel performance: simultaneous multi-dimensional communication and communication-and-computation overlapping. Benchmarking experiments suggest that 3D FFTs of size 128 × 128 × 128 can scale well on such platforms up to 4096 nodes. Our performance results suggest stronger scalability on QCDOC than on IBM BlueGene/L supercomputer.",

keywords = "Ab initio calculation, Blue Matter, FFT, Long-range interaction, Molecular dynamics simulation, Parallel efficiency, Strong scalability",

author = "Bin Fang and Yuefan Deng and Glenn Martyna",

note = "Funding Information: This project is supported by BNL LDRD grant #36930, for molecular dynamics on ultrascalable supercomputer. ",

year = "2007",

month = apr,

day = "15",

doi = "10.1016/j.cpc.2006.12.006",

language = "English (US)",

volume = "176",

pages = "531--538",

journal = "Computer Physics Communications",

issn = "0010-4655",

publisher = "Elsevier B.V.",

number = "8",

}

TY - JOUR

T1 - Performance of the 3D FFT on the 6D network torus QCDOC parallel supercomputer

AU - Fang, Bin

AU - Deng, Yuefan

AU - Martyna, Glenn

N1 - Funding Information: This project is supported by BNL LDRD grant #36930, for molecular dynamics on ultrascalable supercomputer.

PY - 2007/4/15

Y1 - 2007/4/15

N2 - QCDOC is a massively parallel supercomputer with tens of thousands of nodes distributed on a six-dimensional torus network. The 6D structure of the network provides the needed communication resources for many communication-intensive applications. In this paper, we present a parallel algorithm for three-dimensional Fast Fourier Transform and its implementation for a 4096-node QCDOC prototype. Two techniques have been used to increase its parallel performance: simultaneous multi-dimensional communication and communication-and-computation overlapping. Benchmarking experiments suggest that 3D FFTs of size 128 × 128 × 128 can scale well on such platforms up to 4096 nodes. Our performance results suggest stronger scalability on QCDOC than on IBM BlueGene/L supercomputer.

AB - QCDOC is a massively parallel supercomputer with tens of thousands of nodes distributed on a six-dimensional torus network. The 6D structure of the network provides the needed communication resources for many communication-intensive applications. In this paper, we present a parallel algorithm for three-dimensional Fast Fourier Transform and its implementation for a 4096-node QCDOC prototype. Two techniques have been used to increase its parallel performance: simultaneous multi-dimensional communication and communication-and-computation overlapping. Benchmarking experiments suggest that 3D FFTs of size 128 × 128 × 128 can scale well on such platforms up to 4096 nodes. Our performance results suggest stronger scalability on QCDOC than on IBM BlueGene/L supercomputer.

KW - Ab initio calculation

KW - Blue Matter

KW - FFT

KW - Long-range interaction

KW - Molecular dynamics simulation

KW - Parallel efficiency

KW - Strong scalability

UR - http://www.scopus.com/inward/record.url?scp=33947229391&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33947229391&partnerID=8YFLogxK

U2 - 10.1016/j.cpc.2006.12.006

DO - 10.1016/j.cpc.2006.12.006

M3 - Article

AN - SCOPUS:33947229391

SN - 0010-4655

VL - 176

SP - 531

EP - 538

JO - Computer Physics Communications

JF - Computer Physics Communications

IS - 8

ER -

Performance of the 3D FFT on the 6D network torus QCDOC parallel supercomputer

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this