Optimizing parallel performance of streamline visualization for large distributed flow datasets

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

29 Citations (Scopus)

Abstract

Parallel performance has been a challenging topic in streamline visualization for large unstructured flow datasets on parallel distributed-memory computers. It depends strongly on domain partitions. Unsuitable partitions often lead to severe load imbalance and high frequent communications among the domain partitions. To address the problem, we present an approach to flow data partitioning taking account of flow directions and features. Multilevel spectral graph bisection method is employed to reduce communication and synchronization overhead among distributed domains. Edge weights in the corresponding adjacent matrix is defined based on an anisotropic local diffusion operator which assigns strong coupling along flow direction and weak coupling orthogonal to flow. Meanwhile, the distributions of seed points and flow features such as vortex structure are also considered in partitioning so as to obtain good load balance. The experimental results are given to show the feasibility and effectiveness of our method.

Original languageEnglish
Title of host publicationIEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings
Pages87-94
Number of pages8
DOIs
Publication statusPublished - 2008
Externally publishedYes
Event2008 Pacific Visualization Symposium, PacificVis 2008 - Kyoto, Japan
Duration: 2008 Mar 42008 Mar 7

Other

Other2008 Pacific Visualization Symposium, PacificVis 2008
CountryJapan
CityKyoto
Period08/3/408/3/7

Fingerprint

Visualization
Communication
Seed
Synchronization
Vortex flow
Data storage equipment

Keywords

  • Flow clustering
  • Flow visualization
  • Graph partition
  • Load balance
  • Parallel streamline

ASJC Scopus subject areas

  • Computer Graphics and Computer-Aided Design
  • Hardware and Architecture
  • Software

Cite this

Chen, L., & Fujishiro, I. (2008). Optimizing parallel performance of streamline visualization for large distributed flow datasets. In IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings (pp. 87-94). [4475463] https://doi.org/10.1109/PACIFICVIS.2008.4475463

Optimizing parallel performance of streamline visualization for large distributed flow datasets. / Chen, Li; Fujishiro, Issei.

IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings. 2008. p. 87-94 4475463.

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

Chen, L & Fujishiro, I 2008, Optimizing parallel performance of streamline visualization for large distributed flow datasets. in IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings., 4475463, pp. 87-94, 2008 Pacific Visualization Symposium, PacificVis 2008, Kyoto, Japan, 08/3/4. https://doi.org/10.1109/PACIFICVIS.2008.4475463
Chen L, Fujishiro I. Optimizing parallel performance of streamline visualization for large distributed flow datasets. In IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings. 2008. p. 87-94. 4475463 https://doi.org/10.1109/PACIFICVIS.2008.4475463
Chen, Li ; Fujishiro, Issei. / Optimizing parallel performance of streamline visualization for large distributed flow datasets. IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings. 2008. pp. 87-94
@inproceedings{c2eac2edbd44491384094606bcc1b68c,
title = "Optimizing parallel performance of streamline visualization for large distributed flow datasets",
abstract = "Parallel performance has been a challenging topic in streamline visualization for large unstructured flow datasets on parallel distributed-memory computers. It depends strongly on domain partitions. Unsuitable partitions often lead to severe load imbalance and high frequent communications among the domain partitions. To address the problem, we present an approach to flow data partitioning taking account of flow directions and features. Multilevel spectral graph bisection method is employed to reduce communication and synchronization overhead among distributed domains. Edge weights in the corresponding adjacent matrix is defined based on an anisotropic local diffusion operator which assigns strong coupling along flow direction and weak coupling orthogonal to flow. Meanwhile, the distributions of seed points and flow features such as vortex structure are also considered in partitioning so as to obtain good load balance. The experimental results are given to show the feasibility and effectiveness of our method.",
keywords = "Flow clustering, Flow visualization, Graph partition, Load balance, Parallel streamline",
author = "Li Chen and Issei Fujishiro",
year = "2008",
doi = "10.1109/PACIFICVIS.2008.4475463",
language = "English",
isbn = "9781424419661",
pages = "87--94",
booktitle = "IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings",

}

TY - GEN

T1 - Optimizing parallel performance of streamline visualization for large distributed flow datasets

AU - Chen, Li

AU - Fujishiro, Issei

PY - 2008

Y1 - 2008

N2 - Parallel performance has been a challenging topic in streamline visualization for large unstructured flow datasets on parallel distributed-memory computers. It depends strongly on domain partitions. Unsuitable partitions often lead to severe load imbalance and high frequent communications among the domain partitions. To address the problem, we present an approach to flow data partitioning taking account of flow directions and features. Multilevel spectral graph bisection method is employed to reduce communication and synchronization overhead among distributed domains. Edge weights in the corresponding adjacent matrix is defined based on an anisotropic local diffusion operator which assigns strong coupling along flow direction and weak coupling orthogonal to flow. Meanwhile, the distributions of seed points and flow features such as vortex structure are also considered in partitioning so as to obtain good load balance. The experimental results are given to show the feasibility and effectiveness of our method.

AB - Parallel performance has been a challenging topic in streamline visualization for large unstructured flow datasets on parallel distributed-memory computers. It depends strongly on domain partitions. Unsuitable partitions often lead to severe load imbalance and high frequent communications among the domain partitions. To address the problem, we present an approach to flow data partitioning taking account of flow directions and features. Multilevel spectral graph bisection method is employed to reduce communication and synchronization overhead among distributed domains. Edge weights in the corresponding adjacent matrix is defined based on an anisotropic local diffusion operator which assigns strong coupling along flow direction and weak coupling orthogonal to flow. Meanwhile, the distributions of seed points and flow features such as vortex structure are also considered in partitioning so as to obtain good load balance. The experimental results are given to show the feasibility and effectiveness of our method.

KW - Flow clustering

KW - Flow visualization

KW - Graph partition

KW - Load balance

KW - Parallel streamline

UR - http://www.scopus.com/inward/record.url?scp=49749116840&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=49749116840&partnerID=8YFLogxK

U2 - 10.1109/PACIFICVIS.2008.4475463

DO - 10.1109/PACIFICVIS.2008.4475463

M3 - Conference contribution

AN - SCOPUS:49749116840

SN - 9781424419661

SP - 87

EP - 94

BT - IEEE Pacific Visualisation Symposium 2008, PacificVis - Proceedings

ER -