A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning

Dominique Lasalle; George Karypis

doi:10.1109/ICPP.2016.34

A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning

Dominique Lasalle, George Karypis

Computer Science and Engineering

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

35 Scopus citations

Abstract

Graph partitioning is important in distributing workloads on parallel compute systems, computing sparse matrix re-orderings, and designing VLSI circuits. Refinement algorithms are used to improve existing partitionings, and are essential for obtaining high-quality partitionings. Existing parallel refinement algorithms either extract concurrency by sacrificing in terms of quality, or preserve quality by restricting concurrency. In this work we present a new shared-memory parallel algorithm for refining an existing k-way partitioning that can break out of local minima and produce high-quality partitionings. This allows our algorithm to scale well in terms of the number of processing cores and produce clusterings of quality equal to serial algorithms. Our algorithm achieves speedups of 5.7-16.7&-using 24 cores, while exhibiting only 0.52% higher edgecuts than when run serially. This is 6.3x faster and 1.9% better quality than other parallel refinement algorithms which can break out of local minima.

Original language	English (US)
Title of host publication	Proceedings - 45th International Conference on Parallel Processing, ICPP 2016
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	236-241
Number of pages	6
ISBN (Electronic)	9781509028238
DOIs	https://doi.org/10.1109/ICPP.2016.34
State	Published - Sep 21 2016
Event	45th International Conference on Parallel Processing, ICPP 2016 - Philadelphia, United States Duration: Aug 16 2016 → Aug 19 2016

Publication series

Name	Proceedings of the International Conference on Parallel Processing
Volume	2016-September
ISSN (Print)	0190-3918

Other

Other	45th International Conference on Parallel Processing, ICPP 2016
Country/Territory	United States
City	Philadelphia
Period	8/16/16 → 8/19/16

Bibliographical note

Funding Information:
This work was supported in part by NSF (IIS-0905220, OCI-1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788),Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota.

Publisher Copyright:
© 2016 IEEE.

Keywords

Graph partitioning
Local minima
Multithreading

Access

10.1109/ICPP.2016.34

OpenUrl availability

Full text

Cite this

Lasalle, D., & Karypis, G. (2016). A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning. In Proceedings - 45th International Conference on Parallel Processing, ICPP 2016 (pp. 236-241). Article 7573823 (Proceedings of the International Conference on Parallel Processing; Vol. 2016-September). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICPP.2016.34

A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning. / Lasalle, Dominique; Karypis, George.
Proceedings - 45th International Conference on Parallel Processing, ICPP 2016. Institute of Electrical and Electronics Engineers Inc., 2016. p. 236-241 7573823 (Proceedings of the International Conference on Parallel Processing; Vol. 2016-September).

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

Lasalle, D & Karypis, G 2016, A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning. in Proceedings - 45th International Conference on Parallel Processing, ICPP 2016., 7573823, Proceedings of the International Conference on Parallel Processing, vol. 2016-September, Institute of Electrical and Electronics Engineers Inc., pp. 236-241, 45th International Conference on Parallel Processing, ICPP 2016, Philadelphia, United States, 8/16/16. https://doi.org/10.1109/ICPP.2016.34

@inproceedings{d0408ba8fa8049ce8094c24eef55744c,

title = "A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning",

abstract = "Graph partitioning is important in distributing workloads on parallel compute systems, computing sparse matrix re-orderings, and designing VLSI circuits. Refinement algorithms are used to improve existing partitionings, and are essential for obtaining high-quality partitionings. Existing parallel refinement algorithms either extract concurrency by sacrificing in terms of quality, or preserve quality by restricting concurrency. In this work we present a new shared-memory parallel algorithm for refining an existing k-way partitioning that can break out of local minima and produce high-quality partitionings. This allows our algorithm to scale well in terms of the number of processing cores and produce clusterings of quality equal to serial algorithms. Our algorithm achieves speedups of 5.7-16.7&-using 24 cores, while exhibiting only 0.52% higher edgecuts than when run serially. This is 6.3x faster and 1.9% better quality than other parallel refinement algorithms which can break out of local minima.",

keywords = "Graph partitioning, Local minima, Multithreading",

author = "Dominique Lasalle and George Karypis",

note = "Funding Information: This work was supported in part by NSF (IIS-0905220, OCI-1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788),Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota. Publisher Copyright: {\textcopyright} 2016 IEEE.; 45th International Conference on Parallel Processing, ICPP 2016 ; Conference date: 16-08-2016 Through 19-08-2016",

year = "2016",

month = sep,

day = "21",

doi = "10.1109/ICPP.2016.34",

language = "English (US)",

series = "Proceedings of the International Conference on Parallel Processing",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "236--241",

booktitle = "Proceedings - 45th International Conference on Parallel Processing, ICPP 2016",

}

TY - GEN

T1 - A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning

AU - Lasalle, Dominique

AU - Karypis, George

N1 - Funding Information: This work was supported in part by NSF (IIS-0905220, OCI-1048018, CNS-1162405, IIS-1247632, IIP-1414153, IIS-1447788),Army Research Office (W911NF-14-1-0316), Intel Software and Services Group, and the Digital Technology Center at the University of Minnesota. Publisher Copyright: © 2016 IEEE.

PY - 2016/9/21

Y1 - 2016/9/21

N2 - Graph partitioning is important in distributing workloads on parallel compute systems, computing sparse matrix re-orderings, and designing VLSI circuits. Refinement algorithms are used to improve existing partitionings, and are essential for obtaining high-quality partitionings. Existing parallel refinement algorithms either extract concurrency by sacrificing in terms of quality, or preserve quality by restricting concurrency. In this work we present a new shared-memory parallel algorithm for refining an existing k-way partitioning that can break out of local minima and produce high-quality partitionings. This allows our algorithm to scale well in terms of the number of processing cores and produce clusterings of quality equal to serial algorithms. Our algorithm achieves speedups of 5.7-16.7&-using 24 cores, while exhibiting only 0.52% higher edgecuts than when run serially. This is 6.3x faster and 1.9% better quality than other parallel refinement algorithms which can break out of local minima.

AB - Graph partitioning is important in distributing workloads on parallel compute systems, computing sparse matrix re-orderings, and designing VLSI circuits. Refinement algorithms are used to improve existing partitionings, and are essential for obtaining high-quality partitionings. Existing parallel refinement algorithms either extract concurrency by sacrificing in terms of quality, or preserve quality by restricting concurrency. In this work we present a new shared-memory parallel algorithm for refining an existing k-way partitioning that can break out of local minima and produce high-quality partitionings. This allows our algorithm to scale well in terms of the number of processing cores and produce clusterings of quality equal to serial algorithms. Our algorithm achieves speedups of 5.7-16.7&-using 24 cores, while exhibiting only 0.52% higher edgecuts than when run serially. This is 6.3x faster and 1.9% better quality than other parallel refinement algorithms which can break out of local minima.

KW - Graph partitioning

KW - Local minima

KW - Multithreading

UR - http://www.scopus.com/inward/record.url?scp=84990985910&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84990985910&partnerID=8YFLogxK

U2 - 10.1109/ICPP.2016.34

DO - 10.1109/ICPP.2016.34

M3 - Conference contribution

AN - SCOPUS:84990985910

T3 - Proceedings of the International Conference on Parallel Processing

SP - 236

EP - 241

BT - Proceedings - 45th International Conference on Parallel Processing, ICPP 2016

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 45th International Conference on Parallel Processing, ICPP 2016

Y2 - 16 August 2016 through 19 August 2016

ER -

A Parallel Hill-Climbing Refinement Algorithm for Graph Partitioning

Abstract

Publication series

Other

Bibliographical note

Keywords

Access

OpenUrl availability

Other files and links

Fingerprint

Cite this