Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems

Nirmal Kumbhare, Cihan Tunc, Dylan MacHovec, Ali Akoglu, Salim A Hariri, Howard Jay Siegel

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

1 Citation (Scopus)

Abstract

Power-aware scheduling has become a critical research thrust for deploying exascale High Performance Computing (HPC) systems with limited power budget. Time-varying pricing of electricity with respect to the market demand and dynamic HPC workloads can lead to unpredictable operational cost, which complicates the scheduling decisions further. For an oversubscribed HPC system, value based scheduling heuristics have been shown to be a more productive option for scheduling time-constrained tasks over priority and deadline based heuristics. However, oversubscribed HPC systems have higher probability of exceeding the power constraints. Earlier studies on value based heuristics do not take power constraints into account during scheduling decisions. In this study, we propose a methodology for deriving task-specific power-execution time models. These models are derived by interpolating the execution time and power consumption measurements over a configuration space parameterized with pairs of dynamic voltage frequency scaling and forced idleness values. We then propose two power-aware value based heuristics, which utilize those models for power capping the nodes and making resource allocation decisions in an oversubscribed homogeneous HPC system. We compare their performance with traditional value based heuristics under a defined power constraint on a real system using different synthetic traces of scientific computing routines. We show that, as power constraints become tighter, the proposed heuristics significantly outperform earlier heuristics in terms of value earning of the HPC system. We also compare the task completion percentage of proposed heuristics and relate the completion percentage with value earnings of the heuristics.

Original languageEnglish (US)
Title of host publicationProceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages120-130
Number of pages11
ISBN (Electronic)9781538619391
DOIs
StatePublished - Oct 9 2017
Event4th IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017 - Tucson, United States
Duration: Sep 18 2017Sep 22 2017

Other

Other4th IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017
CountryUnited States
CityTucson
Period9/18/179/22/17

Fingerprint

High Performance
Scheduling
Heuristics
Computing
Natural sciences computing
Execution Time
Completion
Percentage
Resource allocation
Costs
Electric power utilization
Electricity
Scientific Computing
Dynamic Performance
Deadline
Configuration Space
Electric potential
Resource Allocation
Power Consumption
Pricing

Keywords

  • heuristics
  • High Performance Computing
  • HPC
  • idle injection
  • power aware
  • power capping
  • scheduling
  • scientific
  • utility
  • value based
  • value function

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Control and Optimization

Cite this

Kumbhare, N., Tunc, C., MacHovec, D., Akoglu, A., Hariri, S. A., & Siegel, H. J. (2017). Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems. In Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017 (pp. 120-130). [8064060] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICCAC.2017.19

Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems. / Kumbhare, Nirmal; Tunc, Cihan; MacHovec, Dylan; Akoglu, Ali; Hariri, Salim A; Siegel, Howard Jay.

Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017. Institute of Electrical and Electronics Engineers Inc., 2017. p. 120-130 8064060.

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

Kumbhare, N, Tunc, C, MacHovec, D, Akoglu, A, Hariri, SA & Siegel, HJ 2017, Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems. in Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017., 8064060, Institute of Electrical and Electronics Engineers Inc., pp. 120-130, 4th IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017, Tucson, United States, 9/18/17. https://doi.org/10.1109/ICCAC.2017.19
Kumbhare N, Tunc C, MacHovec D, Akoglu A, Hariri SA, Siegel HJ. Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems. In Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017. Institute of Electrical and Electronics Engineers Inc. 2017. p. 120-130. 8064060 https://doi.org/10.1109/ICCAC.2017.19
Kumbhare, Nirmal ; Tunc, Cihan ; MacHovec, Dylan ; Akoglu, Ali ; Hariri, Salim A ; Siegel, Howard Jay. / Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems. Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017. Institute of Electrical and Electronics Engineers Inc., 2017. pp. 120-130
@inproceedings{e3f551647bdd4324b9db292d790deaf0,
title = "Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems",
abstract = "Power-aware scheduling has become a critical research thrust for deploying exascale High Performance Computing (HPC) systems with limited power budget. Time-varying pricing of electricity with respect to the market demand and dynamic HPC workloads can lead to unpredictable operational cost, which complicates the scheduling decisions further. For an oversubscribed HPC system, value based scheduling heuristics have been shown to be a more productive option for scheduling time-constrained tasks over priority and deadline based heuristics. However, oversubscribed HPC systems have higher probability of exceeding the power constraints. Earlier studies on value based heuristics do not take power constraints into account during scheduling decisions. In this study, we propose a methodology for deriving task-specific power-execution time models. These models are derived by interpolating the execution time and power consumption measurements over a configuration space parameterized with pairs of dynamic voltage frequency scaling and forced idleness values. We then propose two power-aware value based heuristics, which utilize those models for power capping the nodes and making resource allocation decisions in an oversubscribed homogeneous HPC system. We compare their performance with traditional value based heuristics under a defined power constraint on a real system using different synthetic traces of scientific computing routines. We show that, as power constraints become tighter, the proposed heuristics significantly outperform earlier heuristics in terms of value earning of the HPC system. We also compare the task completion percentage of proposed heuristics and relate the completion percentage with value earnings of the heuristics.",
keywords = "heuristics, High Performance Computing, HPC, idle injection, power aware, power capping, scheduling, scientific, utility, value based, value function",
author = "Nirmal Kumbhare and Cihan Tunc and Dylan MacHovec and Ali Akoglu and Hariri, {Salim A} and Siegel, {Howard Jay}",
year = "2017",
month = "10",
day = "9",
doi = "10.1109/ICCAC.2017.19",
language = "English (US)",
pages = "120--130",
booktitle = "Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - Value Based Scheduling for Oversubscribed Power-Constrained Homogeneous HPC Systems

AU - Kumbhare, Nirmal

AU - Tunc, Cihan

AU - MacHovec, Dylan

AU - Akoglu, Ali

AU - Hariri, Salim A

AU - Siegel, Howard Jay

PY - 2017/10/9

Y1 - 2017/10/9

N2 - Power-aware scheduling has become a critical research thrust for deploying exascale High Performance Computing (HPC) systems with limited power budget. Time-varying pricing of electricity with respect to the market demand and dynamic HPC workloads can lead to unpredictable operational cost, which complicates the scheduling decisions further. For an oversubscribed HPC system, value based scheduling heuristics have been shown to be a more productive option for scheduling time-constrained tasks over priority and deadline based heuristics. However, oversubscribed HPC systems have higher probability of exceeding the power constraints. Earlier studies on value based heuristics do not take power constraints into account during scheduling decisions. In this study, we propose a methodology for deriving task-specific power-execution time models. These models are derived by interpolating the execution time and power consumption measurements over a configuration space parameterized with pairs of dynamic voltage frequency scaling and forced idleness values. We then propose two power-aware value based heuristics, which utilize those models for power capping the nodes and making resource allocation decisions in an oversubscribed homogeneous HPC system. We compare their performance with traditional value based heuristics under a defined power constraint on a real system using different synthetic traces of scientific computing routines. We show that, as power constraints become tighter, the proposed heuristics significantly outperform earlier heuristics in terms of value earning of the HPC system. We also compare the task completion percentage of proposed heuristics and relate the completion percentage with value earnings of the heuristics.

AB - Power-aware scheduling has become a critical research thrust for deploying exascale High Performance Computing (HPC) systems with limited power budget. Time-varying pricing of electricity with respect to the market demand and dynamic HPC workloads can lead to unpredictable operational cost, which complicates the scheduling decisions further. For an oversubscribed HPC system, value based scheduling heuristics have been shown to be a more productive option for scheduling time-constrained tasks over priority and deadline based heuristics. However, oversubscribed HPC systems have higher probability of exceeding the power constraints. Earlier studies on value based heuristics do not take power constraints into account during scheduling decisions. In this study, we propose a methodology for deriving task-specific power-execution time models. These models are derived by interpolating the execution time and power consumption measurements over a configuration space parameterized with pairs of dynamic voltage frequency scaling and forced idleness values. We then propose two power-aware value based heuristics, which utilize those models for power capping the nodes and making resource allocation decisions in an oversubscribed homogeneous HPC system. We compare their performance with traditional value based heuristics under a defined power constraint on a real system using different synthetic traces of scientific computing routines. We show that, as power constraints become tighter, the proposed heuristics significantly outperform earlier heuristics in terms of value earning of the HPC system. We also compare the task completion percentage of proposed heuristics and relate the completion percentage with value earnings of the heuristics.

KW - heuristics

KW - High Performance Computing

KW - HPC

KW - idle injection

KW - power aware

KW - power capping

KW - scheduling

KW - scientific

KW - utility

KW - value based

KW - value function

UR - http://www.scopus.com/inward/record.url?scp=85035311812&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85035311812&partnerID=8YFLogxK

U2 - 10.1109/ICCAC.2017.19

DO - 10.1109/ICCAC.2017.19

M3 - Conference contribution

AN - SCOPUS:85035311812

SP - 120

EP - 130

BT - Proceedings - 2017 IEEE International Conference on Cloud and Autonomic Computing, ICCAC 2017

PB - Institute of Electrical and Electronics Engineers Inc.

ER -