Evaluating random forest classifiers to optimise load balancing of parallel mesh generation

Ananya Gangopadhyay; Paul Bartholomew; Michele Weiland

Evaluating random forest classifiers to optimise load balancing of parallel mesh generation

Ananya Gangopadhyay, Paul Bartholomew, Michele Weiland

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract / Description of output

Mesh generation performance with OpenFOAM’s snappyHexMesh utility is impacted by the decomposition method used for load balancing on HPC systems. Testing shows that 3D hierarchical decomposition with optimal configuration can significantly improve mesh generation runtimes. However, this optimal configuration depends on several factors, such as geometric complexity and mesh resolution requirements. We evaluate the use of random forest classifiers to select optimal subdomain allocations with minimal interaction from the user. Our current implementation achieves a 71% classification accuracy on a limited dataset, showing potential for improvements in future work.

Original language	English
Title of host publication	Parallel CFD 2024
Publication status	Accepted/In press - 8 Jul 2024
Event	35th Parallel CFD International Conference 2024 - Bonn, Germany Duration: 2 Sept 2024 → 4 Sept 2024 https://www.parcfd2024.org/en

Conference

Conference	35th Parallel CFD International Conference 2024
Abbreviated title	ParCFD 2024
Country/Territory	Germany
City	Bonn
Period	2/09/24 → 4/09/24
Internet address	https://www.parcfd2024.org/en

Keywords / Materials (for Non-textual outputs)

computational fluid dynamics
mesh generation
domain decomposition
load balancing
machine learning
parameter optimisation

Access to Document

A Gangopadhyay - ParCFD
For the purpose of open access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission.
Accepted author manuscript, 660 KBLicence: Creative Commons: Attribution (CC-BY)

Cite this

@inproceedings{7fc07a0c5ff54fddaad289c3d1f503fd,

title = "Evaluating random forest classifiers to optimise load balancing of parallel mesh generation",

abstract = "Mesh generation performance with OpenFOAM{\textquoteright}s snappyHexMesh utility is impacted by the decomposition method used for load balancing on HPC systems. Testing shows that 3D hierarchical decomposition with optimal configuration can significantly improve mesh generation runtimes. However, this optimal configuration depends on several factors, such as geometric complexity and mesh resolution requirements. We evaluate the use of random forest classifiers to select optimal subdomain allocations with minimal interaction from the user. Our current implementation achieves a 71% classification accuracy on a limited dataset, showing potential for improvements in future work.",

keywords = "computational fluid dynamics, mesh generation, domain decomposition, load balancing, machine learning, parameter optimisation",

author = "Ananya Gangopadhyay and Paul Bartholomew and Michele Weiland",

year = "2024",

month = jul,

day = "8",

language = "English",

booktitle = "Parallel CFD 2024",

note = "35th Parallel CFD International Conference 2024, ParCFD 2024 ; Conference date: 02-09-2024 Through 04-09-2024",

url = "https://www.parcfd2024.org/en",

}

TY - GEN

T1 - Evaluating random forest classifiers to optimise load balancing of parallel mesh generation

AU - Gangopadhyay, Ananya

AU - Bartholomew, Paul

AU - Weiland, Michele

PY - 2024/7/8

Y1 - 2024/7/8

N2 - Mesh generation performance with OpenFOAM’s snappyHexMesh utility is impacted by the decomposition method used for load balancing on HPC systems. Testing shows that 3D hierarchical decomposition with optimal configuration can significantly improve mesh generation runtimes. However, this optimal configuration depends on several factors, such as geometric complexity and mesh resolution requirements. We evaluate the use of random forest classifiers to select optimal subdomain allocations with minimal interaction from the user. Our current implementation achieves a 71% classification accuracy on a limited dataset, showing potential for improvements in future work.

AB - Mesh generation performance with OpenFOAM’s snappyHexMesh utility is impacted by the decomposition method used for load balancing on HPC systems. Testing shows that 3D hierarchical decomposition with optimal configuration can significantly improve mesh generation runtimes. However, this optimal configuration depends on several factors, such as geometric complexity and mesh resolution requirements. We evaluate the use of random forest classifiers to select optimal subdomain allocations with minimal interaction from the user. Our current implementation achieves a 71% classification accuracy on a limited dataset, showing potential for improvements in future work.

KW - computational fluid dynamics

KW - mesh generation

KW - domain decomposition

KW - load balancing

KW - machine learning

KW - parameter optimisation

M3 - Conference contribution

BT - Parallel CFD 2024

T2 - 35th Parallel CFD International Conference 2024

Y2 - 2 September 2024 through 4 September 2024

ER -

Evaluating random forest classifiers to optimise load balancing of parallel mesh generation

Abstract / Description of output

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this