Asynchronous distributed power iteration with gossip-based normalization

M. Jelasity, Geoffrey Canright, Kenth Engø-Monsen

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

20 Citations (Scopus)

Abstract

The dominant eigenvector of matrices defined by weighted links in overlay networks plays an important role in many peer-to-peer applications. Examples include trust management, importance ranking to support search, and virtual coordinate systems to facilitate managing network proximity. Robust and efficient asynchronous distributed algorithms are known only for the case when the dominant eigenvalue is exactly one. We present a fully distributed algorithm for a more general case: non-negative square matrices that have an arbitrary dominant eigenvalue. The basic idea is that we apply a gossip-based aggregation protocol coupled with an asynchronous iteration algorithm, where the gossip component controls the iteration component. The norm of the resulting vector is an unknown finite constant by default; however, it can optionally be set to any desired constant using a third gossip control component. Through extensive simulation results on artificially generated overlay networks and real web traces we demonstrate the correctness, the performance and the fault tolerance of the protocol.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages514-525
Number of pages12
Volume4641 LNCS
Publication statusPublished - 2007
Event13th International Euro-Par Conference on Parallel Processing, Euro-Par 2007 - Rennes, France
Duration: Aug 28 2007Aug 31 2007

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume4641 LNCS
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other13th International Euro-Par Conference on Parallel Processing, Euro-Par 2007
CountryFrance
CityRennes
Period8/28/078/31/07

Fingerprint

Gossip
Overlay networks
Parallel algorithms
Normalization
Overlay Networks
Distributed Algorithms
Iteration
Network protocols
Asynchronous Iteration
Fault tolerance
User-Computer Interface
Eigenvalues and eigenfunctions
Asynchronous Algorithms
Eigenvalue
Trust Management
Agglomeration
Nonnegative Matrices
Square matrix
Peer to Peer
Fault Tolerance

ASJC Scopus subject areas

  • Computer Science(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Theoretical Computer Science

Cite this

Jelasity, M., Canright, G., & Engø-Monsen, K. (2007). Asynchronous distributed power iteration with gossip-based normalization. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4641 LNCS, pp. 514-525). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 4641 LNCS).

Asynchronous distributed power iteration with gossip-based normalization. / Jelasity, M.; Canright, Geoffrey; Engø-Monsen, Kenth.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 4641 LNCS 2007. p. 514-525 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 4641 LNCS).

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

Jelasity, M, Canright, G & Engø-Monsen, K 2007, Asynchronous distributed power iteration with gossip-based normalization. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 4641 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 4641 LNCS, pp. 514-525, 13th International Euro-Par Conference on Parallel Processing, Euro-Par 2007, Rennes, France, 8/28/07.
Jelasity M, Canright G, Engø-Monsen K. Asynchronous distributed power iteration with gossip-based normalization. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 4641 LNCS. 2007. p. 514-525. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
Jelasity, M. ; Canright, Geoffrey ; Engø-Monsen, Kenth. / Asynchronous distributed power iteration with gossip-based normalization. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 4641 LNCS 2007. pp. 514-525 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{0b8575007cd04d6b8d6259c47e224b0c,
title = "Asynchronous distributed power iteration with gossip-based normalization",
abstract = "The dominant eigenvector of matrices defined by weighted links in overlay networks plays an important role in many peer-to-peer applications. Examples include trust management, importance ranking to support search, and virtual coordinate systems to facilitate managing network proximity. Robust and efficient asynchronous distributed algorithms are known only for the case when the dominant eigenvalue is exactly one. We present a fully distributed algorithm for a more general case: non-negative square matrices that have an arbitrary dominant eigenvalue. The basic idea is that we apply a gossip-based aggregation protocol coupled with an asynchronous iteration algorithm, where the gossip component controls the iteration component. The norm of the resulting vector is an unknown finite constant by default; however, it can optionally be set to any desired constant using a third gossip control component. Through extensive simulation results on artificially generated overlay networks and real web traces we demonstrate the correctness, the performance and the fault tolerance of the protocol.",
author = "M. Jelasity and Geoffrey Canright and Kenth Eng{\o}-Monsen",
year = "2007",
language = "English",
isbn = "9783540744658",
volume = "4641 LNCS",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
pages = "514--525",
booktitle = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

}

TY - GEN

T1 - Asynchronous distributed power iteration with gossip-based normalization

AU - Jelasity, M.

AU - Canright, Geoffrey

AU - Engø-Monsen, Kenth

PY - 2007

Y1 - 2007

N2 - The dominant eigenvector of matrices defined by weighted links in overlay networks plays an important role in many peer-to-peer applications. Examples include trust management, importance ranking to support search, and virtual coordinate systems to facilitate managing network proximity. Robust and efficient asynchronous distributed algorithms are known only for the case when the dominant eigenvalue is exactly one. We present a fully distributed algorithm for a more general case: non-negative square matrices that have an arbitrary dominant eigenvalue. The basic idea is that we apply a gossip-based aggregation protocol coupled with an asynchronous iteration algorithm, where the gossip component controls the iteration component. The norm of the resulting vector is an unknown finite constant by default; however, it can optionally be set to any desired constant using a third gossip control component. Through extensive simulation results on artificially generated overlay networks and real web traces we demonstrate the correctness, the performance and the fault tolerance of the protocol.

AB - The dominant eigenvector of matrices defined by weighted links in overlay networks plays an important role in many peer-to-peer applications. Examples include trust management, importance ranking to support search, and virtual coordinate systems to facilitate managing network proximity. Robust and efficient asynchronous distributed algorithms are known only for the case when the dominant eigenvalue is exactly one. We present a fully distributed algorithm for a more general case: non-negative square matrices that have an arbitrary dominant eigenvalue. The basic idea is that we apply a gossip-based aggregation protocol coupled with an asynchronous iteration algorithm, where the gossip component controls the iteration component. The norm of the resulting vector is an unknown finite constant by default; however, it can optionally be set to any desired constant using a third gossip control component. Through extensive simulation results on artificially generated overlay networks and real web traces we demonstrate the correctness, the performance and the fault tolerance of the protocol.

UR - http://www.scopus.com/inward/record.url?scp=38049176724&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=38049176724&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:38049176724

SN - 9783540744658

VL - 4641 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 514

EP - 525

BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ER -