Adaptive configuration selection for power-constrained heterogeneous systems

Peter E. Bailey, David K Lowenthal, Vignesh Ravi, Barry Rountree, Martin Schulz, Bronis R. De Supinski

Research output: Chapter in Book/Report/Conference proceedingConference contribution

33 Citations (Scopus)

Abstract

As power becomes an increasingly important design factor in high-end supercomputers, future systems will likely operate with power limitations significantly below their peak power specifications. These limitations will be enforced through a combination of software and hardware power policies, which will filter down from the system level to individual nodes. Hardware is already moving in this direction by providing power-capping interfaces to the user. The power/performance trade-off at the node level is critical in maximizing the performance of power-constrained cluster systems, but is also complex because of the many interacting architectural features and accelerators that comprise the hardware configuration of a node. The key to solving this challenge is an accurate power/performance model that will aid in selecting the right configuration from a large set of available configurations. In this paper, we present a novel approach to generate such a model offline using kernel clustering and multivariate linear regression. Our model requires only two iterations to select a configuration, which provides a significant advantage over exhaustive search-based strategies. We apply our model to predict power and performance for different applications using arbitrary configurations, and show that our model, when used with hardware frequency-limiting, selects configurations with significantly higher performance at a given power limit than those chosen by frequency-limiting alone. When applied to a set of 36 computational kernels from a range of applications, our model accurately predicts power and performance, it maintains 91% of optimal performance while meeting power constraints 88% of the time. When the model violates a power constraint, it exceeds the constraint by only 6% in the average case, while simultaneously achieving 54% more performance than an oracle.

Original languageEnglish (US)
Title of host publicationProceedings of the International Conference on Parallel Processing
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages371-380
Number of pages10
Volume2014-November
EditionNovember
DOIs
StatePublished - Nov 13 2014
Event43rd International Conference on Parallel Processing, ICPP 2014 - Minneapolis, United States
Duration: Sep 9 2014Sep 12 2014

Other

Other43rd International Conference on Parallel Processing, ICPP 2014
CountryUnited States
CityMinneapolis
Period9/9/149/12/14

Fingerprint

Constrained Systems
Heterogeneous Systems
Configuration
Hardware
Vertex of a graph
Limiting
Model
kernel
Predict
Exhaustive Search
Supercomputer
Performance Model
Violate
Accelerator
Linear regression
Supercomputers
Large Set
Exceed
High Performance
Particle accelerators

Keywords

  • GPU APU power performance modeling power-constrained

ASJC Scopus subject areas

  • Software
  • Mathematics(all)
  • Hardware and Architecture

Cite this

Bailey, P. E., Lowenthal, D. K., Ravi, V., Rountree, B., Schulz, M., & De Supinski, B. R. (2014). Adaptive configuration selection for power-constrained heterogeneous systems. In Proceedings of the International Conference on Parallel Processing (November ed., Vol. 2014-November, pp. 371-380). [6957246] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICPP.2014.46

Adaptive configuration selection for power-constrained heterogeneous systems. / Bailey, Peter E.; Lowenthal, David K; Ravi, Vignesh; Rountree, Barry; Schulz, Martin; De Supinski, Bronis R.

Proceedings of the International Conference on Parallel Processing. Vol. 2014-November November. ed. Institute of Electrical and Electronics Engineers Inc., 2014. p. 371-380 6957246.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Bailey, PE, Lowenthal, DK, Ravi, V, Rountree, B, Schulz, M & De Supinski, BR 2014, Adaptive configuration selection for power-constrained heterogeneous systems. in Proceedings of the International Conference on Parallel Processing. November edn, vol. 2014-November, 6957246, Institute of Electrical and Electronics Engineers Inc., pp. 371-380, 43rd International Conference on Parallel Processing, ICPP 2014, Minneapolis, United States, 9/9/14. https://doi.org/10.1109/ICPP.2014.46
Bailey PE, Lowenthal DK, Ravi V, Rountree B, Schulz M, De Supinski BR. Adaptive configuration selection for power-constrained heterogeneous systems. In Proceedings of the International Conference on Parallel Processing. November ed. Vol. 2014-November. Institute of Electrical and Electronics Engineers Inc. 2014. p. 371-380. 6957246 https://doi.org/10.1109/ICPP.2014.46
Bailey, Peter E. ; Lowenthal, David K ; Ravi, Vignesh ; Rountree, Barry ; Schulz, Martin ; De Supinski, Bronis R. / Adaptive configuration selection for power-constrained heterogeneous systems. Proceedings of the International Conference on Parallel Processing. Vol. 2014-November November. ed. Institute of Electrical and Electronics Engineers Inc., 2014. pp. 371-380
@inproceedings{a42db808158442e487b3b6017737da8a,
title = "Adaptive configuration selection for power-constrained heterogeneous systems",
abstract = "As power becomes an increasingly important design factor in high-end supercomputers, future systems will likely operate with power limitations significantly below their peak power specifications. These limitations will be enforced through a combination of software and hardware power policies, which will filter down from the system level to individual nodes. Hardware is already moving in this direction by providing power-capping interfaces to the user. The power/performance trade-off at the node level is critical in maximizing the performance of power-constrained cluster systems, but is also complex because of the many interacting architectural features and accelerators that comprise the hardware configuration of a node. The key to solving this challenge is an accurate power/performance model that will aid in selecting the right configuration from a large set of available configurations. In this paper, we present a novel approach to generate such a model offline using kernel clustering and multivariate linear regression. Our model requires only two iterations to select a configuration, which provides a significant advantage over exhaustive search-based strategies. We apply our model to predict power and performance for different applications using arbitrary configurations, and show that our model, when used with hardware frequency-limiting, selects configurations with significantly higher performance at a given power limit than those chosen by frequency-limiting alone. When applied to a set of 36 computational kernels from a range of applications, our model accurately predicts power and performance, it maintains 91{\%} of optimal performance while meeting power constraints 88{\%} of the time. When the model violates a power constraint, it exceeds the constraint by only 6{\%} in the average case, while simultaneously achieving 54{\%} more performance than an oracle.",
keywords = "GPU APU power performance modeling power-constrained",
author = "Bailey, {Peter E.} and Lowenthal, {David K} and Vignesh Ravi and Barry Rountree and Martin Schulz and {De Supinski}, {Bronis R.}",
year = "2014",
month = "11",
day = "13",
doi = "10.1109/ICPP.2014.46",
language = "English (US)",
volume = "2014-November",
pages = "371--380",
booktitle = "Proceedings of the International Conference on Parallel Processing",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
edition = "November",

}

TY - GEN

T1 - Adaptive configuration selection for power-constrained heterogeneous systems

AU - Bailey, Peter E.

AU - Lowenthal, David K

AU - Ravi, Vignesh

AU - Rountree, Barry

AU - Schulz, Martin

AU - De Supinski, Bronis R.

PY - 2014/11/13

Y1 - 2014/11/13

N2 - As power becomes an increasingly important design factor in high-end supercomputers, future systems will likely operate with power limitations significantly below their peak power specifications. These limitations will be enforced through a combination of software and hardware power policies, which will filter down from the system level to individual nodes. Hardware is already moving in this direction by providing power-capping interfaces to the user. The power/performance trade-off at the node level is critical in maximizing the performance of power-constrained cluster systems, but is also complex because of the many interacting architectural features and accelerators that comprise the hardware configuration of a node. The key to solving this challenge is an accurate power/performance model that will aid in selecting the right configuration from a large set of available configurations. In this paper, we present a novel approach to generate such a model offline using kernel clustering and multivariate linear regression. Our model requires only two iterations to select a configuration, which provides a significant advantage over exhaustive search-based strategies. We apply our model to predict power and performance for different applications using arbitrary configurations, and show that our model, when used with hardware frequency-limiting, selects configurations with significantly higher performance at a given power limit than those chosen by frequency-limiting alone. When applied to a set of 36 computational kernels from a range of applications, our model accurately predicts power and performance, it maintains 91% of optimal performance while meeting power constraints 88% of the time. When the model violates a power constraint, it exceeds the constraint by only 6% in the average case, while simultaneously achieving 54% more performance than an oracle.

AB - As power becomes an increasingly important design factor in high-end supercomputers, future systems will likely operate with power limitations significantly below their peak power specifications. These limitations will be enforced through a combination of software and hardware power policies, which will filter down from the system level to individual nodes. Hardware is already moving in this direction by providing power-capping interfaces to the user. The power/performance trade-off at the node level is critical in maximizing the performance of power-constrained cluster systems, but is also complex because of the many interacting architectural features and accelerators that comprise the hardware configuration of a node. The key to solving this challenge is an accurate power/performance model that will aid in selecting the right configuration from a large set of available configurations. In this paper, we present a novel approach to generate such a model offline using kernel clustering and multivariate linear regression. Our model requires only two iterations to select a configuration, which provides a significant advantage over exhaustive search-based strategies. We apply our model to predict power and performance for different applications using arbitrary configurations, and show that our model, when used with hardware frequency-limiting, selects configurations with significantly higher performance at a given power limit than those chosen by frequency-limiting alone. When applied to a set of 36 computational kernels from a range of applications, our model accurately predicts power and performance, it maintains 91% of optimal performance while meeting power constraints 88% of the time. When the model violates a power constraint, it exceeds the constraint by only 6% in the average case, while simultaneously achieving 54% more performance than an oracle.

KW - GPU APU power performance modeling power-constrained

UR - http://www.scopus.com/inward/record.url?scp=84932631259&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84932631259&partnerID=8YFLogxK

U2 - 10.1109/ICPP.2014.46

DO - 10.1109/ICPP.2014.46

M3 - Conference contribution

VL - 2014-November

SP - 371

EP - 380

BT - Proceedings of the International Conference on Parallel Processing

PB - Institute of Electrical and Electronics Engineers Inc.

ER -