Merged requests for better performance and productivity in multithreaded OpenSHMEM

Swen Boehm; Swaroop Pophale; Matthew B. Baker; Manjunath Gorentla Venkata

doi:10.1007/978-3-319-73814-7_3

Merged requests for better performance and productivity in multithreaded OpenSHMEM

Swen Boehm, Swaroop Pophale, Matthew B. Baker, Manjunath Gorentla Venkata

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

1 Scopus citations

Abstract

A merged request is a handle representing a group of Remote Memory Access (RMA), Atomic or Collective operations. The merged request can be created either by combining multiple outstanding merged request handles or using the same merged request handle for additional operations. We show that introducing such simple yet powerful semantics in OpenSHMEM provides many productivity and performance advantages. In this paper, we first introduce the interfaces and semantics for creating and using merged request handles. Then, we demonstrate with a merge request that we can achieve better performance characteristics in multithreaded OpenSHMEM application. Particularly, we show one can achieve higher message rate, a higher bandwidth for smaller message, and better computation-communication overlap. Further, we use merged request to realize multithreaded collectives, where multiple threads co-operate to complete the collective operation. Our experimental results show that in a multithreaded OpenSHMEM program, the merged request based RMA operations achieve over 100 Million Messages Per Second (MMPS). It achieves over 10 MMPS compared to 4.5 MMPS with default RMA operations in a single threaded environment. Also, we achieve higher bandwidth for smaller message sizes, close to 100% overlap, and reduce the latency by 60%.

Original language	English
Title of host publication	OpenSHMEM and Related Technologies
Subtitle of host publication	Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers
Editors	Manjunath Gorentla Venkata, Neena Imam, Swaroop Pophale
Publisher	Springer Verlag
Pages	35-49
Number of pages	15
ISBN (Print)	9783319738130
DOIs	https://doi.org/10.1007/978-3-319-73814-7_3
State	Published - 2018
Event	4th Workshop on OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence, OpenSHMEM 2017 - Annapolis, United States Duration: Aug 7 2017 → Aug 9 2017

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10679 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	4th Workshop on OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence, OpenSHMEM 2017
Country/Territory	United States
City	Annapolis
Period	08/7/17 → 08/9/17

Funding

This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).

Keywords

Interoperability
PGAS
Shared memory

Access to Document

10.1007/978-3-319-73814-7_3

Cite this

Boehm, S., Pophale, S., Baker, M. B., & Venkata, M. G. (2018). Merged requests for better performance and productivity in multithreaded OpenSHMEM. In M. Gorentla Venkata, N. Imam, & S. Pophale (Eds.), OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers (pp. 35-49). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10679 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-73814-7_3

Boehm, Swen ; Pophale, Swaroop ; Baker, Matthew B. et al. / Merged requests for better performance and productivity in multithreaded OpenSHMEM. OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers. editor / Manjunath Gorentla Venkata ; Neena Imam ; Swaroop Pophale. Springer Verlag, 2018. pp. 35-49 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{5c7ef59301ee452d831a660e202804e8,

title = "Merged requests for better performance and productivity in multithreaded OpenSHMEM",

abstract = "A merged request is a handle representing a group of Remote Memory Access (RMA), Atomic or Collective operations. The merged request can be created either by combining multiple outstanding merged request handles or using the same merged request handle for additional operations. We show that introducing such simple yet powerful semantics in OpenSHMEM provides many productivity and performance advantages. In this paper, we first introduce the interfaces and semantics for creating and using merged request handles. Then, we demonstrate with a merge request that we can achieve better performance characteristics in multithreaded OpenSHMEM application. Particularly, we show one can achieve higher message rate, a higher bandwidth for smaller message, and better computation-communication overlap. Further, we use merged request to realize multithreaded collectives, where multiple threads co-operate to complete the collective operation. Our experimental results show that in a multithreaded OpenSHMEM program, the merged request based RMA operations achieve over 100 Million Messages Per Second (MMPS). It achieves over 10 MMPS compared to 4.5 MMPS with default RMA operations in a single threaded environment. Also, we achieve higher bandwidth for smaller message sizes, close to 100% overlap, and reduce the latency by 60%.",

keywords = "Interoperability, PGAS, Shared memory",

author = "Swen Boehm and Swaroop Pophale and Baker, {Matthew B.} and Venkata, {Manjunath Gorentla}",

note = "Publisher Copyright: {\textcopyright} Springer International Publishing AG 2018.; 4th Workshop on OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence, OpenSHMEM 2017 ; Conference date: 07-08-2017 Through 09-08-2017",

year = "2018",

doi = "10.1007/978-3-319-73814-7_3",

language = "English",

isbn = "9783319738130",

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

publisher = "Springer Verlag",

pages = "35--49",

editor = "{Gorentla Venkata}, Manjunath and Neena Imam and Swaroop Pophale",

booktitle = "OpenSHMEM and Related Technologies",

}

Boehm, S , Pophale, S, Baker, MB & Venkata, MG 2018, Merged requests for better performance and productivity in multithreaded OpenSHMEM. in M Gorentla Venkata, N Imam & S Pophale (eds), OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10679 LNCS, Springer Verlag, pp. 35-49, 4th Workshop on OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence, OpenSHMEM 2017, Annapolis, United States, 08/7/17. https://doi.org/10.1007/978-3-319-73814-7_3

Merged requests for better performance and productivity in multithreaded OpenSHMEM. / Boehm, Swen ; Pophale, Swaroop; Baker, Matthew B. et al.
OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers. ed. / Manjunath Gorentla Venkata; Neena Imam; Swaroop Pophale. Springer Verlag, 2018. p. 35-49 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10679 LNCS).

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

TY - GEN

T1 - Merged requests for better performance and productivity in multithreaded OpenSHMEM

AU - Boehm, Swen

AU - Pophale, Swaroop

AU - Baker, Matthew B.

AU - Venkata, Manjunath Gorentla

N1 - Publisher Copyright: © Springer International Publishing AG 2018.

PY - 2018

Y1 - 2018

N2 - A merged request is a handle representing a group of Remote Memory Access (RMA), Atomic or Collective operations. The merged request can be created either by combining multiple outstanding merged request handles or using the same merged request handle for additional operations. We show that introducing such simple yet powerful semantics in OpenSHMEM provides many productivity and performance advantages. In this paper, we first introduce the interfaces and semantics for creating and using merged request handles. Then, we demonstrate with a merge request that we can achieve better performance characteristics in multithreaded OpenSHMEM application. Particularly, we show one can achieve higher message rate, a higher bandwidth for smaller message, and better computation-communication overlap. Further, we use merged request to realize multithreaded collectives, where multiple threads co-operate to complete the collective operation. Our experimental results show that in a multithreaded OpenSHMEM program, the merged request based RMA operations achieve over 100 Million Messages Per Second (MMPS). It achieves over 10 MMPS compared to 4.5 MMPS with default RMA operations in a single threaded environment. Also, we achieve higher bandwidth for smaller message sizes, close to 100% overlap, and reduce the latency by 60%.

AB - A merged request is a handle representing a group of Remote Memory Access (RMA), Atomic or Collective operations. The merged request can be created either by combining multiple outstanding merged request handles or using the same merged request handle for additional operations. We show that introducing such simple yet powerful semantics in OpenSHMEM provides many productivity and performance advantages. In this paper, we first introduce the interfaces and semantics for creating and using merged request handles. Then, we demonstrate with a merge request that we can achieve better performance characteristics in multithreaded OpenSHMEM application. Particularly, we show one can achieve higher message rate, a higher bandwidth for smaller message, and better computation-communication overlap. Further, we use merged request to realize multithreaded collectives, where multiple threads co-operate to complete the collective operation. Our experimental results show that in a multithreaded OpenSHMEM program, the merged request based RMA operations achieve over 100 Million Messages Per Second (MMPS). It achieves over 10 MMPS compared to 4.5 MMPS with default RMA operations in a single threaded environment. Also, we achieve higher bandwidth for smaller message sizes, close to 100% overlap, and reduce the latency by 60%.

KW - Interoperability

KW - PGAS

KW - Shared memory

UR - http://www.scopus.com/inward/record.url?scp=85041500952&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-73814-7_3

DO - 10.1007/978-3-319-73814-7_3

M3 - Conference contribution

AN - SCOPUS:85041500952

SN - 9783319738130

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

SP - 35

EP - 49

BT - OpenSHMEM and Related Technologies

A2 - Gorentla Venkata, Manjunath

A2 - Imam, Neena

A2 - Pophale, Swaroop

PB - Springer Verlag

T2 - 4th Workshop on OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence, OpenSHMEM 2017

Y2 - 7 August 2017 through 9 August 2017

ER -

Boehm S , Pophale S, Baker MB, Venkata MG. Merged requests for better performance and productivity in multithreaded OpenSHMEM. In Gorentla Venkata M, Imam N, Pophale S, editors, OpenSHMEM and Related Technologies: Big Compute and Big Data Convergence - 4th Workshop, OpenSHMEM 2017, Revised Selected Papers. Springer Verlag. 2018. p. 35-49. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-73814-7_3

Merged requests for better performance and productivity in multithreaded OpenSHMEM

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this