Scalability Issues in FFT Computation

Alan Ayala; Stanimire Tomov; Miroslav Stoyanov; Jack Dongarra

doi:10.1007/978-3-030-86359-3_21

Scalability Issues in FFT Computation

Alan Ayala, Stanimire Tomov, Miroslav Stoyanov, Jack Dongarra

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

7 Scopus citations

Abstract

The fast Fourier transform (FFT), is one the most important tools in mathematics, and it is widely required by several applications of science and engineering. State-of-the-art parallel implementations of the FFT algorithm, based on Cooley-Tukey developments, are known to be communication-bound, which causes critical issues when scaling the computational and architectural capabilities. In this paper, we study the main performance bottleneck of FFT computations on hybrid CPU and GPU systems at large-scale. We provide numerical simulations and potential acceleration techniques that can be easily integrated into FFT distributed libraries. We present different experiments on performance scalability and runtime analysis on the world’s most powerful supercomputers today: Summit, using up to 6,144 NVIDIA V100 GPUs, and Fugaku, using more than one million Fujitsu A64FX cores.

Original language	English
Title of host publication	Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings
Editors	Victor Malyshkin
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	279-287
Number of pages	9
ISBN (Print)	9783030863586
DOIs	https://doi.org/10.1007/978-3-030-86359-3_21
State	Published - 2021
Event	16th International Conference on Parallel Computing Technologies, PaCT 2021 - Kaliningrad, Russian Federation Duration: Sep 13 2021 → Sep 18 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12942 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	16th International Conference on Parallel Computing Technologies, PaCT 2021
Country/Territory	Russian Federation
City	Kaliningrad
Period	09/13/21 → 09/18/21

Funding

This research was supported by the Exascale Computing Project (ECP), Project Number: 17-SC-20-SC, a collaborative effort of two DOE organizations (the Office of Science and the National Nuclear Security Administration) responsible for the planning and preparation of a capable exascale ecosystem.

Funders	Funder number
DOE organizations
Office of Science
National Nuclear Security Administration

Keywords

Hybrid systems
Parallel FFT
Scalability

Access to Document

10.1007/978-3-030-86359-3_21

Cite this

Ayala, A., Tomov, S., Stoyanov, M., & Dongarra, J. (2021). Scalability Issues in FFT Computation. In V. Malyshkin (Ed.), Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings (pp. 279-287). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12942 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-86359-3_21

Ayala, Alan ; Tomov, Stanimire ; Stoyanov, Miroslav et al. / Scalability Issues in FFT Computation. Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings. editor / Victor Malyshkin. Springer Science and Business Media Deutschland GmbH, 2021. pp. 279-287 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e54874378a0e45f29635d8b7637e3cd2,

title = "Scalability Issues in FFT Computation",

abstract = "The fast Fourier transform (FFT), is one the most important tools in mathematics, and it is widely required by several applications of science and engineering. State-of-the-art parallel implementations of the FFT algorithm, based on Cooley-Tukey developments, are known to be communication-bound, which causes critical issues when scaling the computational and architectural capabilities. In this paper, we study the main performance bottleneck of FFT computations on hybrid CPU and GPU systems at large-scale. We provide numerical simulations and potential acceleration techniques that can be easily integrated into FFT distributed libraries. We present different experiments on performance scalability and runtime analysis on the world{\textquoteright}s most powerful supercomputers today: Summit, using up to 6,144 NVIDIA V100 GPUs, and Fugaku, using more than one million Fujitsu A64FX cores.",

keywords = "Hybrid systems, Parallel FFT, Scalability",

author = "Alan Ayala and Stanimire Tomov and Miroslav Stoyanov and Jack Dongarra",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 16th International Conference on Parallel Computing Technologies, PaCT 2021 ; Conference date: 13-09-2021 Through 18-09-2021",

year = "2021",

doi = "10.1007/978-3-030-86359-3_21",

language = "English",

isbn = "9783030863586",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "279--287",

editor = "Victor Malyshkin",

booktitle = "Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings",

}

Ayala, A, Tomov, S, Stoyanov, M & Dongarra, J 2021, Scalability Issues in FFT Computation. in V Malyshkin (ed.), Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12942 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 279-287, 16th International Conference on Parallel Computing Technologies, PaCT 2021, Kaliningrad, Russian Federation, 09/13/21. https://doi.org/10.1007/978-3-030-86359-3_21

Scalability Issues in FFT Computation. / Ayala, Alan; Tomov, Stanimire; Stoyanov, Miroslav et al.
Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings. ed. / Victor Malyshkin. Springer Science and Business Media Deutschland GmbH, 2021. p. 279-287 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12942 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Scalability Issues in FFT Computation

AU - Ayala, Alan

AU - Tomov, Stanimire

AU - Stoyanov, Miroslav

AU - Dongarra, Jack

PY - 2021

Y1 - 2021

N2 - The fast Fourier transform (FFT), is one the most important tools in mathematics, and it is widely required by several applications of science and engineering. State-of-the-art parallel implementations of the FFT algorithm, based on Cooley-Tukey developments, are known to be communication-bound, which causes critical issues when scaling the computational and architectural capabilities. In this paper, we study the main performance bottleneck of FFT computations on hybrid CPU and GPU systems at large-scale. We provide numerical simulations and potential acceleration techniques that can be easily integrated into FFT distributed libraries. We present different experiments on performance scalability and runtime analysis on the world’s most powerful supercomputers today: Summit, using up to 6,144 NVIDIA V100 GPUs, and Fugaku, using more than one million Fujitsu A64FX cores.

AB - The fast Fourier transform (FFT), is one the most important tools in mathematics, and it is widely required by several applications of science and engineering. State-of-the-art parallel implementations of the FFT algorithm, based on Cooley-Tukey developments, are known to be communication-bound, which causes critical issues when scaling the computational and architectural capabilities. In this paper, we study the main performance bottleneck of FFT computations on hybrid CPU and GPU systems at large-scale. We provide numerical simulations and potential acceleration techniques that can be easily integrated into FFT distributed libraries. We present different experiments on performance scalability and runtime analysis on the world’s most powerful supercomputers today: Summit, using up to 6,144 NVIDIA V100 GPUs, and Fugaku, using more than one million Fujitsu A64FX cores.

KW - Hybrid systems

KW - Parallel FFT

KW - Scalability

UR - http://www.scopus.com/inward/record.url?scp=85115336216&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-86359-3_21

DO - 10.1007/978-3-030-86359-3_21

M3 - Conference contribution

AN - SCOPUS:85115336216

SN - 9783030863586

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 279

EP - 287

BT - Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings

A2 - Malyshkin, Victor

PB - Springer Science and Business Media Deutschland GmbH

T2 - 16th International Conference on Parallel Computing Technologies, PaCT 2021

Y2 - 13 September 2021 through 18 September 2021

ER -

Ayala A, Tomov S, Stoyanov M, Dongarra J. Scalability Issues in FFT Computation. In Malyshkin V, editor, Parallel Computing Technologies - 16th International Conference, PaCT 2021, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 279-287. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-86359-3_21

Scalability Issues in FFT Computation

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this