Hierarchical multiscale simulations for data center cooling

Surya Pratap Vanka, Jayati Athavale, Minami Yoda, Yogendra Joshi

Research output: Contribution to journalConference article

Abstract

Accurate computation of the cooling of data centers is a challenging multiscale problem with scales spanning several orders of magnitude. Several hierarchical levels of models have been developed in the literature. These range from network based models to more complex turbulent CFD models based on Reynolds-averaged formulations. In this paper, we summarize these approaches, and describe some techniques to significantly accelerate the CFD models. Three techniques, namely the multigrid method, multilevel refinement, and GPU based parallel computing, are described.

Original languageEnglish (US)
JournalInternational Conference on Computational Methods for Thermal Problems
Issue number217349
StatePublished - Jan 1 2016
Event4th International Conference on Computational Methods for Thermal Problems, THERMACOMP 2016 - Atlanta, United States
Duration: Jul 6 2016Jul 8 2016

Fingerprint

Multiscale Simulation
Data Center
Cooling
Multiscale Problems
Computational fluid dynamics
Multigrid Method
Parallel Computing
Accelerate
Refinement
Parallel processing systems
Model
Model-based
Formulation
Range of data

Keywords

  • Data center cooling
  • GPU
  • Hierarchical models
  • Multigrid method

ASJC Scopus subject areas

  • Fluid Flow and Transfer Processes
  • Computational Mathematics
  • Numerical Analysis

Cite this

Hierarchical multiscale simulations for data center cooling. / Vanka, Surya Pratap; Athavale, Jayati; Yoda, Minami; Joshi, Yogendra.

In: International Conference on Computational Methods for Thermal Problems, No. 217349, 01.01.2016.

Research output: Contribution to journalConference article

@article{cb7e77c74e7842ad8fc00db6c78981b7,
title = "Hierarchical multiscale simulations for data center cooling",
abstract = "Accurate computation of the cooling of data centers is a challenging multiscale problem with scales spanning several orders of magnitude. Several hierarchical levels of models have been developed in the literature. These range from network based models to more complex turbulent CFD models based on Reynolds-averaged formulations. In this paper, we summarize these approaches, and describe some techniques to significantly accelerate the CFD models. Three techniques, namely the multigrid method, multilevel refinement, and GPU based parallel computing, are described.",
keywords = "Data center cooling, GPU, Hierarchical models, Multigrid method",
author = "Vanka, {Surya Pratap} and Jayati Athavale and Minami Yoda and Yogendra Joshi",
year = "2016",
month = "1",
day = "1",
language = "English (US)",
journal = "International Conference on Computational Methods for Thermal Problems",
issn = "2305-5995",
number = "217349",

}

TY - JOUR

T1 - Hierarchical multiscale simulations for data center cooling

AU - Vanka, Surya Pratap

AU - Athavale, Jayati

AU - Yoda, Minami

AU - Joshi, Yogendra

PY - 2016/1/1

Y1 - 2016/1/1

N2 - Accurate computation of the cooling of data centers is a challenging multiscale problem with scales spanning several orders of magnitude. Several hierarchical levels of models have been developed in the literature. These range from network based models to more complex turbulent CFD models based on Reynolds-averaged formulations. In this paper, we summarize these approaches, and describe some techniques to significantly accelerate the CFD models. Three techniques, namely the multigrid method, multilevel refinement, and GPU based parallel computing, are described.

AB - Accurate computation of the cooling of data centers is a challenging multiscale problem with scales spanning several orders of magnitude. Several hierarchical levels of models have been developed in the literature. These range from network based models to more complex turbulent CFD models based on Reynolds-averaged formulations. In this paper, we summarize these approaches, and describe some techniques to significantly accelerate the CFD models. Three techniques, namely the multigrid method, multilevel refinement, and GPU based parallel computing, are described.

KW - Data center cooling

KW - GPU

KW - Hierarchical models

KW - Multigrid method

UR - http://www.scopus.com/inward/record.url?scp=85053665448&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85053665448&partnerID=8YFLogxK

M3 - Conference article

JO - International Conference on Computational Methods for Thermal Problems

JF - International Conference on Computational Methods for Thermal Problems

SN - 2305-5995

IS - 217349

ER -