ytopt: Autotuning Scientific Applications for Energy Efficiency at Large Scales

Xingfu Wu; Prasanna Balaprakash; Michael Kruse; Jaehoon Koo; Brice Videau; Paul Hovland; Valerie Taylor; Brad Geltz; Siddhartha Jana; Mary Hall

doi:10.1002/cpe.8322

ytopt: Autotuning Scientific Applications for Energy Efficiency at Large Scales

Xingfu Wu, Prasanna Balaprakash, Michael Kruse, Jaehoon Koo, Brice Videau, Paul Hovland, Valerie Taylor, Brad Geltz, Siddhartha Jana, Mary Hall

CCSD Planning, Partnerships, Operations

Research output: Contribution to journal › Article › peer-review

Abstract

As we enter the exascale computing era, efficiently utilizing power and optimizing the performance of scientific applications under power and energy constraints has become critical and challenging. We propose a low-overhead autotuning framework to autotune performance and energy for various hybrid MPI/OpenMP scientific applications at large scales and to explore the tradeoffs between application runtime and power/energy for energy efficient application execution, then use this framework to autotune four ECP proxy applications—XSBench, AMG, SWFFT, and SW4lite. Our approach uses Bayesian optimization with a Random Forest surrogate model to effectively search parameter spaces with up to 6 million different configurations on two large-scale HPC production systems, Theta at Argonne National Laboratory and Summit at Oak Ridge National Laboratory. The experimental results show that our autotuning framework at large scales has low overhead and achieves good scalability. Using the proposed autotuning framework to identify the best configurations, we achieve up to 91.59% performance improvement, up to 21.2% energy savings, and up to 37.84% EDP (energy delay product) improvement on up to 4096 nodes.

Original language	English
Article number	e8322
Journal	Concurrency and Computation: Practice and Experience
Volume	37
Issue number	1
DOIs	https://doi.org/10.1002/cpe.8322
State	Published - Jan 10 2025

Funding

Funding: This work was supported in part by DOE ECP PROTEAS-TUNE, and in part by DOE ASCR RAPIDS2 and OASIS. We acknowledge the Argonne Leadership Computing Facility (ALCF) for use of Cray XC40 Theta under ALCF projects EE-ECP and Intel, and the Oak Ridge Leadership Computing Facility for use of Summit under the projects CSC383, MED106, and AST136. We also acknowledge Adrian Pope at ALCF for providing the SWFFT problem sizes. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract number DE-AC02-06CH11357, SciDAC-RAPIDS-OASIS;ECP. Development of the GEOPM software package has been partially funded through contract B609815 with Argonne National Laboratory. This work was supported in part by DOE ECP PROTEAS-TUNE, and in part by DOE ASCR RAPIDS2 and OASIS. We acknowledge the Argonne Leadership Computing Facility (ALCF) for use of Cray XC40 Theta under ALCF projects EE-ECP and Intel, and the Oak Ridge Leadership Computing Facility for use of Summit under the projects CSC383, MED106, and AST136. We also acknowledge Adrian Pope at ALCF for providing the SWFFT problem sizes. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract number DE-AC02-06CH11357. Development of the GEOPM software package has been partially funded through contract B609815 with Argonne National Laboratory. This work was supported in part by DOE ECP PROTEAS\u2010TUNE, and in part by DOE ASCR RAPIDS2 and OASIS. We acknowledge the Argonne Leadership Computing Facility (ALCF) for use of Cray XC40 Theta under ALCF projects EE\u2010ECP and Intel, and the Oak Ridge Leadership Computing Facility for use of Summit under the projects CSC383, MED106, and AST136. We also acknowledge Adrian Pope at ALCF for providing the SWFFT problem sizes. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract number DE\u2010AC02\u201006CH11357. Development of the GEOPM software package has been partially funded through contract B609815 with Argonne National Laboratory. This work was supported in part by DOE ECP PROTEAS\u2010TUNE, and in part by DOE ASCR RAPIDS2 and OASIS. We acknowledge the Argonne Leadership Computing Facility (ALCF) for use of Cray XC40 Theta under ALCF projects EE\u2010ECP and Intel, and the Oak Ridge Leadership Computing Facility for use of Summit under the projects CSC383, MED106, and AST136. We also acknowledge Adrian Pope at ALCF for providing the SWFFT problem sizes. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract number DE\u2010AC02\u201006CH11357, SciDAC\u2010RAPIDS\u2010OASIS;ECP. Development of the GEOPM software package has been partially funded through contract B609815 with Argonne National Laboratory. Funding:

Keywords

Bayesian optimization
autotuning
energy efficiency
hybrid MPI/OpenMP applications
performance optimization

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1002/cpe.8322

Cite this

@article{118cd61e25d843ee9bee8988e85a8dac,

title = "ytopt: Autotuning Scientific Applications for Energy Efficiency at Large Scales",

abstract = "As we enter the exascale computing era, efficiently utilizing power and optimizing the performance of scientific applications under power and energy constraints has become critical and challenging. We propose a low-overhead autotuning framework to autotune performance and energy for various hybrid MPI/OpenMP scientific applications at large scales and to explore the tradeoffs between application runtime and power/energy for energy efficient application execution, then use this framework to autotune four ECP proxy applications—XSBench, AMG, SWFFT, and SW4lite. Our approach uses Bayesian optimization with a Random Forest surrogate model to effectively search parameter spaces with up to 6 million different configurations on two large-scale HPC production systems, Theta at Argonne National Laboratory and Summit at Oak Ridge National Laboratory. The experimental results show that our autotuning framework at large scales has low overhead and achieves good scalability. Using the proposed autotuning framework to identify the best configurations, we achieve up to 91.59% performance improvement, up to 21.2% energy savings, and up to 37.84% EDP (energy delay product) improvement on up to 4096 nodes.",

keywords = "Bayesian optimization, autotuning, energy efficiency, hybrid MPI/OpenMP applications, performance optimization",

author = "Xingfu Wu and Prasanna Balaprakash and Michael Kruse and Jaehoon Koo and Brice Videau and Paul Hovland and Valerie Taylor and Brad Geltz and Siddhartha Jana and Mary Hall",

note = "Publisher Copyright: {\textcopyright} 2024 UChicago Argonne, LLC and The Author(s). Concurrency and Computation: Practice and Experience published by John Wiley & Sons Ltd.",

year = "2025",

month = jan,

day = "10",

doi = "10.1002/cpe.8322",

language = "English",

volume = "37",

journal = "Concurrency and Computation: Practice and Experience",

issn = "1532-0626",

publisher = "wiley",

number = "1",

}

TY - JOUR

T1 - ytopt

T2 - Autotuning Scientific Applications for Energy Efficiency at Large Scales

AU - Wu, Xingfu

AU - Balaprakash, Prasanna

AU - Kruse, Michael

AU - Koo, Jaehoon

AU - Videau, Brice

AU - Hovland, Paul

AU - Taylor, Valerie

AU - Geltz, Brad

AU - Jana, Siddhartha

AU - Hall, Mary

PY - 2025/1/10

Y1 - 2025/1/10

N2 - As we enter the exascale computing era, efficiently utilizing power and optimizing the performance of scientific applications under power and energy constraints has become critical and challenging. We propose a low-overhead autotuning framework to autotune performance and energy for various hybrid MPI/OpenMP scientific applications at large scales and to explore the tradeoffs between application runtime and power/energy for energy efficient application execution, then use this framework to autotune four ECP proxy applications—XSBench, AMG, SWFFT, and SW4lite. Our approach uses Bayesian optimization with a Random Forest surrogate model to effectively search parameter spaces with up to 6 million different configurations on two large-scale HPC production systems, Theta at Argonne National Laboratory and Summit at Oak Ridge National Laboratory. The experimental results show that our autotuning framework at large scales has low overhead and achieves good scalability. Using the proposed autotuning framework to identify the best configurations, we achieve up to 91.59% performance improvement, up to 21.2% energy savings, and up to 37.84% EDP (energy delay product) improvement on up to 4096 nodes.

AB - As we enter the exascale computing era, efficiently utilizing power and optimizing the performance of scientific applications under power and energy constraints has become critical and challenging. We propose a low-overhead autotuning framework to autotune performance and energy for various hybrid MPI/OpenMP scientific applications at large scales and to explore the tradeoffs between application runtime and power/energy for energy efficient application execution, then use this framework to autotune four ECP proxy applications—XSBench, AMG, SWFFT, and SW4lite. Our approach uses Bayesian optimization with a Random Forest surrogate model to effectively search parameter spaces with up to 6 million different configurations on two large-scale HPC production systems, Theta at Argonne National Laboratory and Summit at Oak Ridge National Laboratory. The experimental results show that our autotuning framework at large scales has low overhead and achieves good scalability. Using the proposed autotuning framework to identify the best configurations, we achieve up to 91.59% performance improvement, up to 21.2% energy savings, and up to 37.84% EDP (energy delay product) improvement on up to 4096 nodes.

KW - Bayesian optimization

KW - autotuning

KW - energy efficiency

KW - hybrid MPI/OpenMP applications

KW - performance optimization

UR - http://www.scopus.com/inward/record.url?scp=85208078660&partnerID=8YFLogxK

U2 - 10.1002/cpe.8322

DO - 10.1002/cpe.8322

M3 - Article

AN - SCOPUS:85208078660

SN - 1532-0626

VL - 37

JO - Concurrency and Computation: Practice and Experience

JF - Concurrency and Computation: Practice and Experience

IS - 1

M1 - e8322

ER -

ytopt: Autotuning Scientific Applications for Energy Efficiency at Large Scales

Abstract

Funding

Keywords

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this