Dynamic power management of a computer with self power-managed components

Maryam Triki; Yanzhi Wang; Ahmed C. Ammari; Massoud Pedram

doi:10.1007/978-3-642-36157-9_22

Dynamic power management of a computer with self power-managed components

Maryam Triki, Yanzhi Wang, Ahmed C. Ammari, Massoud Pedram

Research output: Chapter in Book/Entry/Poem › Conference contribution

1 Scopus citations

Abstract

This paper presents a Dynamic Power Management (DPM) framework based on reinforcement learning (RL) technique which aims to save power in an Energy-Managed Computer (EMC) system with self power-managed components. The proposed online adaptive DPM technique consists of two layers: component-level and system-level global power manager (GPM). The component-level PM policy is pre-specified and fixed whereas the system-level global PM employs temporal difference learning on Semi-Markov Decision Process (SMDP) for model-free RL, and it is specifically optimized for a multitype application framework. Experiments show that that the proposed HPM scheme enhances power savings considerably while maintaining a good performance level. In comparison with other reference systems, the proposed RL DPM approach performs well under various workloads, can simultaneously consider power and performance and achieves a wide and deep powerperformance tradeoff curves.

Original language	English (US)
Title of host publication	Integrated Circuit and System Design
Subtitle of host publication	Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers
Pages	215-224
Number of pages	10
DOIs	https://doi.org/10.1007/978-3-642-36157-9_22
State	Published - 2013
Externally published	Yes
Event	22nd International Workshop on Power and Timing Modeling, Optimization and Simulation, PATMOS 2012 - Newcastle upon Tyne, United Kingdom Duration: Sep 4 2012 → Sep 6 2012

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	7606 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Other

Other	22nd International Workshop on Power and Timing Modeling, Optimization and Simulation, PATMOS 2012
Country/Territory	United Kingdom
City	Newcastle upon Tyne
Period	9/4/12 → 9/6/12

Keywords

Dynamic Power Management (DPM)
Power optimization
Reinforcement Learning (RL)

ASJC Scopus subject areas

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-642-36157-9_22

Cite this

Triki, M., Wang, Y., Ammari, A. C., & Pedram, M. (2013). Dynamic power management of a computer with self power-managed components. In Integrated Circuit and System Design: Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers (pp. 215-224). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7606 LNCS). https://doi.org/10.1007/978-3-642-36157-9_22

Dynamic power management of a computer with self power-managed components. / Triki, Maryam; Wang, Yanzhi; Ammari, Ahmed C. et al.
Integrated Circuit and System Design: Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers. 2013. p. 215-224 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7606 LNCS).

Research output: Chapter in Book/Entry/Poem › Conference contribution

Triki, M, Wang, Y, Ammari, AC & Pedram, M 2013, Dynamic power management of a computer with self power-managed components. in Integrated Circuit and System Design: Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 7606 LNCS, pp. 215-224, 22nd International Workshop on Power and Timing Modeling, Optimization and Simulation, PATMOS 2012, Newcastle upon Tyne, United Kingdom, 9/4/12. https://doi.org/10.1007/978-3-642-36157-9_22

Triki M, Wang Y, Ammari AC, Pedram M. Dynamic power management of a computer with self power-managed components. In Integrated Circuit and System Design: Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers. 2013. p. 215-224. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-642-36157-9_22

Triki, Maryam ; Wang, Yanzhi ; Ammari, Ahmed C. et al. / Dynamic power management of a computer with self power-managed components. Integrated Circuit and System Design: Power and Timing Modeling, Optimization and Simulation - 22nd International Workshop, PATMOS 2012, Revised Selected Papers. 2013. pp. 215-224 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{4a4140683e7042c698350f98b3e2f411,

title = "Dynamic power management of a computer with self power-managed components",

abstract = "This paper presents a Dynamic Power Management (DPM) framework based on reinforcement learning (RL) technique which aims to save power in an Energy-Managed Computer (EMC) system with self power-managed components. The proposed online adaptive DPM technique consists of two layers: component-level and system-level global power manager (GPM). The component-level PM policy is pre-specified and fixed whereas the system-level global PM employs temporal difference learning on Semi-Markov Decision Process (SMDP) for model-free RL, and it is specifically optimized for a multitype application framework. Experiments show that that the proposed HPM scheme enhances power savings considerably while maintaining a good performance level. In comparison with other reference systems, the proposed RL DPM approach performs well under various workloads, can simultaneously consider power and performance and achieves a wide and deep powerperformance tradeoff curves.",

keywords = "Dynamic Power Management (DPM), Power optimization, Reinforcement Learning (RL)",

author = "Maryam Triki and Yanzhi Wang and Ammari, {Ahmed C.} and Massoud Pedram",

year = "2013",

doi = "10.1007/978-3-642-36157-9_22",

language = "English (US)",

isbn = "9783642361562",

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

pages = "215--224",

booktitle = "Integrated Circuit and System Design",

note = "22nd International Workshop on Power and Timing Modeling, Optimization and Simulation, PATMOS 2012 ; Conference date: 04-09-2012 Through 06-09-2012",

}

TY - GEN

T1 - Dynamic power management of a computer with self power-managed components

AU - Triki, Maryam

AU - Wang, Yanzhi

AU - Ammari, Ahmed C.

AU - Pedram, Massoud

PY - 2013

Y1 - 2013

N2 - This paper presents a Dynamic Power Management (DPM) framework based on reinforcement learning (RL) technique which aims to save power in an Energy-Managed Computer (EMC) system with self power-managed components. The proposed online adaptive DPM technique consists of two layers: component-level and system-level global power manager (GPM). The component-level PM policy is pre-specified and fixed whereas the system-level global PM employs temporal difference learning on Semi-Markov Decision Process (SMDP) for model-free RL, and it is specifically optimized for a multitype application framework. Experiments show that that the proposed HPM scheme enhances power savings considerably while maintaining a good performance level. In comparison with other reference systems, the proposed RL DPM approach performs well under various workloads, can simultaneously consider power and performance and achieves a wide and deep powerperformance tradeoff curves.

AB - This paper presents a Dynamic Power Management (DPM) framework based on reinforcement learning (RL) technique which aims to save power in an Energy-Managed Computer (EMC) system with self power-managed components. The proposed online adaptive DPM technique consists of two layers: component-level and system-level global power manager (GPM). The component-level PM policy is pre-specified and fixed whereas the system-level global PM employs temporal difference learning on Semi-Markov Decision Process (SMDP) for model-free RL, and it is specifically optimized for a multitype application framework. Experiments show that that the proposed HPM scheme enhances power savings considerably while maintaining a good performance level. In comparison with other reference systems, the proposed RL DPM approach performs well under various workloads, can simultaneously consider power and performance and achieves a wide and deep powerperformance tradeoff curves.

KW - Dynamic Power Management (DPM)

KW - Power optimization

KW - Reinforcement Learning (RL)

UR - http://www.scopus.com/inward/record.url?scp=84893411711&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84893411711&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-36157-9_22

DO - 10.1007/978-3-642-36157-9_22

M3 - Conference contribution

AN - SCOPUS:84893411711

SN - 9783642361562

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

SP - 215

EP - 224

BT - Integrated Circuit and System Design

T2 - 22nd International Workshop on Power and Timing Modeling, Optimization and Simulation, PATMOS 2012

Y2 - 4 September 2012 through 6 September 2012

ER -

Dynamic power management of a computer with self power-managed components

Abstract

Publication series

Other

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this