Design of fast collective communication functions on clustered workstations with ethernet and myrinet

Dongyoung Kim, Dongseung Kim

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

Abstract

In cluster computing, current communication functions under MPI library are not well optimized. Especially, the performance is worse if there are multiple sources and/or destinations involved, which are the cases of collective communication. Our algorithms uses multidimensional factorization and pairwise exchange communication/dissemination methods to improve the performance. They deliver better performance than previous algorithms such as ring, recursive doubling and dissemination algorithms. Experimental results show the improvement of 50% or so over MPICH version 1.2.6 on a Linux cluster.

Original languageEnglish
Title of host publicationHigh-Performance Computing - 6th International Symposium, ISHPC 2005 and First International Workshop on Advanced Low Power Systems, ALPS 2006, Revised Selected Papers
PublisherSpringer Verlag
Pages105-116
Number of pages12
ISBN (Print)3540777032, 9783540777038
DOIs
Publication statusPublished - 2008
Event6th International Symposium on High Performance Computing, ISHPC 2005 and 1st International Workshop on Advanced Low Power Systems, ALPS 2006 - Nara, Japan
Duration: 2005 Sept 72005 Sept 9

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume4759 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference6th International Symposium on High Performance Computing, ISHPC 2005 and 1st International Workshop on Advanced Low Power Systems, ALPS 2006
Country/TerritoryJapan
CityNara
Period05/9/705/9/9

Keywords

  • Broadcast
  • Cluster computing
  • MPI
  • Message passing communication

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Design of fast collective communication functions on clustered workstations with ethernet and myrinet'. Together they form a unique fingerprint.

Cite this