User co-scheduling for MPI+OpenMP applications using OpenMP semantics

Antoine Capra; Patrick Carribault; Jean Baptiste Besnard; Allen D. Malony; Marc Pérache; Julien Jaeger

doi:10.1007/978-3-319-65578-9_14

User co-scheduling for MPI+OpenMP applications using OpenMP semantics

Antoine Capra, Patrick Carribault, Jean Baptiste Besnard, Allen D. Malony, Marc Pérache, Julien Jaeger

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

Abstract

The evolution of parallel architectures towards machines with many-core processors and high node-level concurrency is putting an end to the pure-MPI programming model. Simulations codes must expose multiple levels of parallelisms inside and between nodes, combining different programming models (e.g., MPI+X), to productively use current and future supercomputers. MPI+OpenMP is a common hybridization approach. However, recent evolutions in the OpenMP standard presents options for how OpenMP tasking constructs might be used when mixing fine-grained computation and communications. Various approaches are discussed and compared in this context. Advantages and limitations of the approaches are detailed, including potential improvements to OpenMP in order ease both the integration and progress of MPI calls. These methods are applied to a representative stencil code and demonstrate improvements on the overall execution time as a result of more efficient mixing of MPI and OpenMP.

Original language	English
Title of host publication	Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings
Editors	Bronis R. de Supinski, Barbara M. Chapman, Christian Terboven, Matthias S. Muller, Stephen L. Olivier
Publisher	Springer Verlag
Pages	203-216
Number of pages	14
ISBN (Print)	9783319655772
DOIs	https://doi.org/10.1007/978-3-319-65578-9_14
Publication status	Published - 1 Jan 2017
Externally published	Yes
Event	13th International Workshop on OpenMP, IWOMP 2017 - Stony Brook, United States Duration: 20 Sept 2017 → 22 Sept 2017

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10468 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	13th International Workshop on OpenMP, IWOMP 2017
Country/Territory	United States
City	Stony Brook
Period	20/09/17 → 22/09/17

Access to Document

10.1007/978-3-319-65578-9_14

Cite this

Capra, A., Carribault, P., Besnard, J. B., Malony, A. D., Pérache, M., & Jaeger, J. (2017). User co-scheduling for MPI+OpenMP applications using OpenMP semantics. In B. R. de Supinski, B. M. Chapman, C. Terboven, M. S. Muller, & S. L. Olivier (Eds.), Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings (pp. 203-216). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10468 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-65578-9_14

Capra, Antoine ; Carribault, Patrick ; Besnard, Jean Baptiste et al. / User co-scheduling for MPI+OpenMP applications using OpenMP semantics. Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings. editor / Bronis R. de Supinski ; Barbara M. Chapman ; Christian Terboven ; Matthias S. Muller ; Stephen L. Olivier. Springer Verlag, 2017. pp. 203-216 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{a07292206de3428b89185d5dbecf226d,

title = "User co-scheduling for MPI+OpenMP applications using OpenMP semantics",

abstract = "The evolution of parallel architectures towards machines with many-core processors and high node-level concurrency is putting an end to the pure-MPI programming model. Simulations codes must expose multiple levels of parallelisms inside and between nodes, combining different programming models (e.g., MPI+X), to productively use current and future supercomputers. MPI+OpenMP is a common hybridization approach. However, recent evolutions in the OpenMP standard presents options for how OpenMP tasking constructs might be used when mixing fine-grained computation and communications. Various approaches are discussed and compared in this context. Advantages and limitations of the approaches are detailed, including potential improvements to OpenMP in order ease both the integration and progress of MPI calls. These methods are applied to a representative stencil code and demonstrate improvements on the overall execution time as a result of more efficient mixing of MPI and OpenMP.",

author = "Antoine Capra and Patrick Carribault and Besnard, \{Jean Baptiste\} and Malony, \{Allen D.\} and Marc P{\'e}rache and Julien Jaeger",

note = "Publisher Copyright: {\textcopyright} Springer International Publishing AG 2017.; 13th International Workshop on OpenMP, IWOMP 2017 ; Conference date: 20-09-2017 Through 22-09-2017",

year = "2017",

month = jan,

day = "1",

doi = "10.1007/978-3-319-65578-9\_14",

language = "English",

isbn = "9783319655772",

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

publisher = "Springer Verlag",

pages = "203--216",

editor = "\{de Supinski\}, \{Bronis R.\} and Chapman, \{Barbara M.\} and Christian Terboven and Muller, \{Matthias S.\} and Olivier, \{Stephen L.\}",

booktitle = "Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings",

}

Capra, A, Carribault, P, Besnard, JB, Malony, AD, Pérache, M & Jaeger, J 2017, User co-scheduling for MPI+OpenMP applications using OpenMP semantics. in BR de Supinski, BM Chapman, C Terboven, MS Muller & SL Olivier (eds), Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10468 LNCS, Springer Verlag, pp. 203-216, 13th International Workshop on OpenMP, IWOMP 2017, Stony Brook, United States, 20/09/17. https://doi.org/10.1007/978-3-319-65578-9_14

User co-scheduling for MPI+OpenMP applications using OpenMP semantics. / Capra, Antoine; Carribault, Patrick; Besnard, Jean Baptiste et al.
Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings. ed. / Bronis R. de Supinski; Barbara M. Chapman; Christian Terboven; Matthias S. Muller; Stephen L. Olivier. Springer Verlag, 2017. p. 203-216 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10468 LNCS).

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

TY - GEN

T1 - User co-scheduling for MPI+OpenMP applications using OpenMP semantics

AU - Capra, Antoine

AU - Carribault, Patrick

AU - Besnard, Jean Baptiste

AU - Malony, Allen D.

AU - Pérache, Marc

AU - Jaeger, Julien

N1 - Publisher Copyright: © Springer International Publishing AG 2017.

PY - 2017/1/1

Y1 - 2017/1/1

N2 - The evolution of parallel architectures towards machines with many-core processors and high node-level concurrency is putting an end to the pure-MPI programming model. Simulations codes must expose multiple levels of parallelisms inside and between nodes, combining different programming models (e.g., MPI+X), to productively use current and future supercomputers. MPI+OpenMP is a common hybridization approach. However, recent evolutions in the OpenMP standard presents options for how OpenMP tasking constructs might be used when mixing fine-grained computation and communications. Various approaches are discussed and compared in this context. Advantages and limitations of the approaches are detailed, including potential improvements to OpenMP in order ease both the integration and progress of MPI calls. These methods are applied to a representative stencil code and demonstrate improvements on the overall execution time as a result of more efficient mixing of MPI and OpenMP.

AB - The evolution of parallel architectures towards machines with many-core processors and high node-level concurrency is putting an end to the pure-MPI programming model. Simulations codes must expose multiple levels of parallelisms inside and between nodes, combining different programming models (e.g., MPI+X), to productively use current and future supercomputers. MPI+OpenMP is a common hybridization approach. However, recent evolutions in the OpenMP standard presents options for how OpenMP tasking constructs might be used when mixing fine-grained computation and communications. Various approaches are discussed and compared in this context. Advantages and limitations of the approaches are detailed, including potential improvements to OpenMP in order ease both the integration and progress of MPI calls. These methods are applied to a representative stencil code and demonstrate improvements on the overall execution time as a result of more efficient mixing of MPI and OpenMP.

U2 - 10.1007/978-3-319-65578-9_14

DO - 10.1007/978-3-319-65578-9_14

M3 - Conference contribution

AN - SCOPUS:85029348342

SN - 9783319655772

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

SP - 203

EP - 216

BT - Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings

A2 - de Supinski, Bronis R.

A2 - Chapman, Barbara M.

A2 - Terboven, Christian

A2 - Muller, Matthias S.

A2 - Olivier, Stephen L.

PB - Springer Verlag

T2 - 13th International Workshop on OpenMP, IWOMP 2017

Y2 - 20 September 2017 through 22 September 2017

ER -

Capra A, Carribault P, Besnard JB, Malony AD, Pérache M, Jaeger J. User co-scheduling for MPI+OpenMP applications using OpenMP semantics. In de Supinski BR, Chapman BM, Terboven C, Muller MS, Olivier SL, editors, Scaling OpenMP for Exascale Performance and Portability - 13th International Workshop on OpenMP, IWOMP 2017, Proceedings. Springer Verlag. 2017. p. 203-216. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-65578-9_14

User co-scheduling for MPI+OpenMP applications using OpenMP semantics

Abstract

Publication series

Conference

Access to Document

Fingerprint

Cite this