Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications

Romain Pereira; Adrien Roussel; Patrick Carribault; Thierry Gautier

doi:10.1007/978-3-030-85262-7_14

Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications

Romain Pereira
, Adrien Roussel
, Patrick Carribault
, Thierry Gautier

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

Abstract

While task-based programming, such as OpenMP, is a promising solution to exploit large HPC compute nodes, it has to be mixed with data communications like MPI. However, performance or even more thread progression may depend on the underlying runtime implementations. In this paper, we focus on enhancing the application performance when an OpenMP task blocks inside MPI communications. This technique requires no additional effort on the application developers. It relies on an online task re-ordering strategy that aims at running first tasks that are sending data to other processes. We evaluate our approach on a Cholesky factorization and show that we gain around 19% of execution time on an Intel Skylake compute nodes machine - each node having two 24-core processors.

Original language	English
Title of host publication	OpenMP
Subtitle of host publication	Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings
Editors	Simon McIntosh-Smith, Bronis R. de Supinski, Jannis Klinkenberg
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	197-210
Number of pages	14
ISBN (Print)	9783030852610
DOIs	https://doi.org/10.1007/978-3-030-85262-7_14
Publication status	Published - 1 Jan 2021
Externally published	Yes
Event	17th International Workshop on OpenMP, IWOMP 2021 - Bristol, United Kingdom Duration: 14 Sept 2021 → 16 Sept 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12870 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	17th International Workshop on OpenMP, IWOMP 2021
Country/Territory	United Kingdom
City	Bristol
Period	14/09/21 → 16/09/21

Keywords

Asynchronism
MPI+OpenMP
Scheduling
Task

Access to Document

10.1007/978-3-030-85262-7_14

Cite this

Pereira, R., Roussel, A., Carribault, P., & Gautier, T. (2021). Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications. In S. McIntosh-Smith, B. R. de Supinski, & J. Klinkenberg (Eds.), OpenMP: Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings (pp. 197-210). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12870 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-85262-7_14

Pereira, Romain ; Roussel, Adrien ; Carribault, Patrick et al. / Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications. OpenMP: Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings. editor / Simon McIntosh-Smith ; Bronis R. de Supinski ; Jannis Klinkenberg. Springer Science and Business Media Deutschland GmbH, 2021. pp. 197-210 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e0aa485783074319b01e514c71a6303b,

title = "Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications",

abstract = "While task-based programming, such as OpenMP, is a promising solution to exploit large HPC compute nodes, it has to be mixed with data communications like MPI. However, performance or even more thread progression may depend on the underlying runtime implementations. In this paper, we focus on enhancing the application performance when an OpenMP task blocks inside MPI communications. This technique requires no additional effort on the application developers. It relies on an online task re-ordering strategy that aims at running first tasks that are sending data to other processes. We evaluate our approach on a Cholesky factorization and show that we gain around 19\% of execution time on an Intel Skylake compute nodes machine - each node having two 24-core processors.",

keywords = "Asynchronism, MPI+OpenMP, Scheduling, Task",

author = "Romain Pereira and Adrien Roussel and Patrick Carribault and Thierry Gautier",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 17th International Workshop on OpenMP, IWOMP 2021 ; Conference date: 14-09-2021 Through 16-09-2021",

year = "2021",

month = jan,

day = "1",

doi = "10.1007/978-3-030-85262-7\_14",

language = "English",

isbn = "9783030852610",

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 = "197--210",

editor = "Simon McIntosh-Smith and \{de Supinski\}, \{Bronis R.\} and Jannis Klinkenberg",

booktitle = "OpenMP",

}

Pereira, R, Roussel, A, Carribault, P & Gautier, T 2021, Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications. in S McIntosh-Smith, BR de Supinski & J Klinkenberg (eds), OpenMP: Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12870 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 197-210, 17th International Workshop on OpenMP, IWOMP 2021, Bristol, United Kingdom, 14/09/21. https://doi.org/10.1007/978-3-030-85262-7_14

Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications. / Pereira, Romain; Roussel, Adrien; Carribault, Patrick et al.
OpenMP: Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings. ed. / Simon McIntosh-Smith; Bronis R. de Supinski; Jannis Klinkenberg. Springer Science and Business Media Deutschland GmbH, 2021. p. 197-210 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12870 LNCS).

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

TY - GEN

T1 - Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications

AU - Pereira, Romain

AU - Roussel, Adrien

AU - Carribault, Patrick

AU - Gautier, Thierry

PY - 2021/1/1

Y1 - 2021/1/1

N2 - While task-based programming, such as OpenMP, is a promising solution to exploit large HPC compute nodes, it has to be mixed with data communications like MPI. However, performance or even more thread progression may depend on the underlying runtime implementations. In this paper, we focus on enhancing the application performance when an OpenMP task blocks inside MPI communications. This technique requires no additional effort on the application developers. It relies on an online task re-ordering strategy that aims at running first tasks that are sending data to other processes. We evaluate our approach on a Cholesky factorization and show that we gain around 19% of execution time on an Intel Skylake compute nodes machine - each node having two 24-core processors.

AB - While task-based programming, such as OpenMP, is a promising solution to exploit large HPC compute nodes, it has to be mixed with data communications like MPI. However, performance or even more thread progression may depend on the underlying runtime implementations. In this paper, we focus on enhancing the application performance when an OpenMP task blocks inside MPI communications. This technique requires no additional effort on the application developers. It relies on an online task re-ordering strategy that aims at running first tasks that are sending data to other processes. We evaluate our approach on a Cholesky factorization and show that we gain around 19% of execution time on an Intel Skylake compute nodes machine - each node having two 24-core processors.

KW - Asynchronism

KW - MPI+OpenMP

KW - Scheduling

KW - Task

UR - https://www.scopus.com/pages/publications/85115446608

U2 - 10.1007/978-3-030-85262-7_14

DO - 10.1007/978-3-030-85262-7_14

M3 - Conference contribution

AN - SCOPUS:85115446608

SN - 9783030852610

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

SP - 197

EP - 210

BT - OpenMP

A2 - McIntosh-Smith, Simon

A2 - de Supinski, Bronis R.

A2 - Klinkenberg, Jannis

PB - Springer Science and Business Media Deutschland GmbH

T2 - 17th International Workshop on OpenMP, IWOMP 2021

Y2 - 14 September 2021 through 16 September 2021

ER -

Pereira R, Roussel A, Carribault P, Gautier T. Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications. In McIntosh-Smith S, de Supinski BR, Klinkenberg J, editors, OpenMP: Enabling Massive Node-Level Parallelism - 17th International Workshop on OpenMP, IWOMP 2021, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 197-210. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-85262-7_14

Communication-Aware Task Scheduling Strategy in Hybrid MPI+OpenMP Applications

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this