Improving uintah's scalability through the use of portable kokkos-based data parallel tasks

John K. Holmen; Alan Humphrey; Daniel Sunderland; Martin Berzins

doi:10.1145/3093338.3093388

Improving uintah's scalability through the use of portable kokkos-based data parallel tasks

John K. Holmen
, Alan Humphrey
, Daniel Sunderland
, Martin Berzins

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

12 Scopus citations

Abstract

.e University of Utah's Carbon Capture Multidisciplinary Simulation Center (CCMSC) is using the Uintah Computational Framework to predict performance of a 1000 MWe ultra-supercritical clean coal boiler. .e center aims to utilize the Intel Xeon Phi-based DOE systems, .eta and Aurora, through the Aurora Early Science Program by using the Kokkos C++ library to enable node-level performance portability. .is paper describes infrastructure advancements and portability improvements made possible by the integration of Kokkos within Uintah. .is integration marks a step towards consolidating Uintah's MPI+P.reads and MPI+CUDA hybrid parallelism approaches into a single MPI+Kokkos approach. Scalability results are presented that compare serial and data parallel task execution models for a challenging radiative heat transfer calculation, central to the center's predictive boiler simulations. .ese results demonstrate both good strong-scaling characteristics to 256 Knights Landing (KNL) processors on the NSF Stampede system, and show the KNL-based calculation to compete with prior GPU-based results for the same calculation.

Original language	English
Title of host publication	PEARC 2017 - Practice and Experience in Advanced Research Computing 2017
Subtitle of host publication	Sustainability, Success and Impact
Publisher	Association for Computing Machinery
ISBN (Electronic)	9781450352727
DOIs	https://doi.org/10.1145/3093338.3093388
State	Published - Jul 9 2017
Externally published	Yes
Event	2017 Practice and Experience in Advanced Research Computing, PEARC 2017 - New Orleans, United States Duration: Jul 9 2017 → Jul 13 2017

Publication series

Name	ACM International Conference Proceeding Series
Volume	Part F128771

Conference

Conference	2017 Practice and Experience in Advanced Research Computing, PEARC 2017
Country/Territory	United States
City	New Orleans
Period	07/9/17 → 07/13/17

Keywords

Hybrid Parallelism
Knights Landing
Kokkos
MIC
Many-Core
Parallel
Portability
Radiation Modeling
Reverse Monte-Carlo Ray Tracing
Scalability
Stampede
Uintah
Xeon Phi

Access to Document

10.1145/3093338.3093388

Cite this

Holmen, J. K., Humphrey, A., Sunderland, D., & Berzins, M. (2017). Improving uintah's scalability through the use of portable kokkos-based data parallel tasks. In PEARC 2017 - Practice and Experience in Advanced Research Computing 2017: Sustainability, Success and Impact Article a27 (ACM International Conference Proceeding Series; Vol. Part F128771). Association for Computing Machinery. https://doi.org/10.1145/3093338.3093388

@inproceedings{f5c59b6dffd142cf9d2878aea788e55f,

title = "Improving uintah's scalability through the use of portable kokkos-based data parallel tasks",

abstract = ".e University of Utah's Carbon Capture Multidisciplinary Simulation Center (CCMSC) is using the Uintah Computational Framework to predict performance of a 1000 MWe ultra-supercritical clean coal boiler. .e center aims to utilize the Intel Xeon Phi-based DOE systems, .eta and Aurora, through the Aurora Early Science Program by using the Kokkos C++ library to enable node-level performance portability. .is paper describes infrastructure advancements and portability improvements made possible by the integration of Kokkos within Uintah. .is integration marks a step towards consolidating Uintah's MPI+P.reads and MPI+CUDA hybrid parallelism approaches into a single MPI+Kokkos approach. Scalability results are presented that compare serial and data parallel task execution models for a challenging radiative heat transfer calculation, central to the center's predictive boiler simulations. .ese results demonstrate both good strong-scaling characteristics to 256 Knights Landing (KNL) processors on the NSF Stampede system, and show the KNL-based calculation to compete with prior GPU-based results for the same calculation.",

keywords = "Hybrid Parallelism, Knights Landing, Kokkos, MIC, Many-Core, Parallel, Portability, Radiation Modeling, Reverse Monte-Carlo Ray Tracing, Scalability, Stampede, Uintah, Xeon Phi",

author = "Holmen, \{John K.\} and Alan Humphrey and Daniel Sunderland and Martin Berzins",

note = "Publisher Copyright: {\textcopyright} 2017 ACM.; 2017 Practice and Experience in Advanced Research Computing, PEARC 2017 ; Conference date: 09-07-2017 Through 13-07-2017",

year = "2017",

month = jul,

day = "9",

doi = "10.1145/3093338.3093388",

language = "English",

series = "ACM International Conference Proceeding Series",

publisher = "Association for Computing Machinery",

booktitle = "PEARC 2017 - Practice and Experience in Advanced Research Computing 2017",

}

Holmen, JK, Humphrey, A, Sunderland, D & Berzins, M 2017, Improving uintah's scalability through the use of portable kokkos-based data parallel tasks. in PEARC 2017 - Practice and Experience in Advanced Research Computing 2017: Sustainability, Success and Impact., a27, ACM International Conference Proceeding Series, vol. Part F128771, Association for Computing Machinery, 2017 Practice and Experience in Advanced Research Computing, PEARC 2017, New Orleans, United States, 07/9/17. https://doi.org/10.1145/3093338.3093388

Improving uintah's scalability through the use of portable kokkos-based data parallel tasks. / Holmen, John K.; Humphrey, Alan; Sunderland, Daniel et al.
PEARC 2017 - Practice and Experience in Advanced Research Computing 2017: Sustainability, Success and Impact. Association for Computing Machinery, 2017. a27 (ACM International Conference Proceeding Series; Vol. Part F128771).

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

TY - GEN

T1 - Improving uintah's scalability through the use of portable kokkos-based data parallel tasks

AU - Holmen, John K.

AU - Humphrey, Alan

AU - Sunderland, Daniel

AU - Berzins, Martin

PY - 2017/7/9

Y1 - 2017/7/9

N2 - .e University of Utah's Carbon Capture Multidisciplinary Simulation Center (CCMSC) is using the Uintah Computational Framework to predict performance of a 1000 MWe ultra-supercritical clean coal boiler. .e center aims to utilize the Intel Xeon Phi-based DOE systems, .eta and Aurora, through the Aurora Early Science Program by using the Kokkos C++ library to enable node-level performance portability. .is paper describes infrastructure advancements and portability improvements made possible by the integration of Kokkos within Uintah. .is integration marks a step towards consolidating Uintah's MPI+P.reads and MPI+CUDA hybrid parallelism approaches into a single MPI+Kokkos approach. Scalability results are presented that compare serial and data parallel task execution models for a challenging radiative heat transfer calculation, central to the center's predictive boiler simulations. .ese results demonstrate both good strong-scaling characteristics to 256 Knights Landing (KNL) processors on the NSF Stampede system, and show the KNL-based calculation to compete with prior GPU-based results for the same calculation.

AB - .e University of Utah's Carbon Capture Multidisciplinary Simulation Center (CCMSC) is using the Uintah Computational Framework to predict performance of a 1000 MWe ultra-supercritical clean coal boiler. .e center aims to utilize the Intel Xeon Phi-based DOE systems, .eta and Aurora, through the Aurora Early Science Program by using the Kokkos C++ library to enable node-level performance portability. .is paper describes infrastructure advancements and portability improvements made possible by the integration of Kokkos within Uintah. .is integration marks a step towards consolidating Uintah's MPI+P.reads and MPI+CUDA hybrid parallelism approaches into a single MPI+Kokkos approach. Scalability results are presented that compare serial and data parallel task execution models for a challenging radiative heat transfer calculation, central to the center's predictive boiler simulations. .ese results demonstrate both good strong-scaling characteristics to 256 Knights Landing (KNL) processors on the NSF Stampede system, and show the KNL-based calculation to compete with prior GPU-based results for the same calculation.

KW - Hybrid Parallelism

KW - Knights Landing

KW - Kokkos

KW - MIC

KW - Many-Core

KW - Parallel

KW - Portability

KW - Radiation Modeling

KW - Reverse Monte-Carlo Ray Tracing

KW - Scalability

KW - Stampede

KW - Uintah

KW - Xeon Phi

UR - https://www.scopus.com/pages/publications/85025832467

U2 - 10.1145/3093338.3093388

DO - 10.1145/3093338.3093388

M3 - Conference contribution

AN - SCOPUS:85025832467

T3 - ACM International Conference Proceeding Series

BT - PEARC 2017 - Practice and Experience in Advanced Research Computing 2017

PB - Association for Computing Machinery

T2 - 2017 Practice and Experience in Advanced Research Computing, PEARC 2017

Y2 - 9 July 2017 through 13 July 2017

ER -

Holmen JK, Humphrey A, Sunderland D, Berzins M. Improving uintah's scalability through the use of portable kokkos-based data parallel tasks. In PEARC 2017 - Practice and Experience in Advanced Research Computing 2017: Sustainability, Success and Impact. Association for Computing Machinery. 2017. a27. (ACM International Conference Proceeding Series). doi: 10.1145/3093338.3093388

Improving uintah's scalability through the use of portable kokkos-based data parallel tasks

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this