A network extend method of OpenMPI

  • Yaohui Hu*
  • , Limin Xiao
  • , Ruan Li
  • , Yongnan Li
  • , Tian Li
  • *Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper is mainly to present a method to extend the network under Open MPI parallel environment. We walk through the Modular Component Architecture (MCA) and three core elements: framework, component and module comprised in the MCA. After an analysis of the point-to-point communication based on the network components in Open MPI, we present a method to extend a network by creating a network component under Byte Transfer Layer (BTL) framework, implementing the basic functions the component must have, configuring the network component to adjust to different architecture platforms and building the whole project to make the new component work in the framework. Finally we successfully extend the network of Open MPI with Hyper Transport (HT). By running a parallel program on the newly extended Open MPI based on HT. We prove the network extend method is accessible.

Original languageEnglish
Title of host publicationProceedings - 2010 3rd IEEE International Conference on Computer Science and Information Technology, ICCSIT 2010
Pages366-370
Number of pages5
DOIs
StatePublished - 2010
Event2010 3rd IEEE International Conference on Computer Science and Information Technology, ICCSIT 2010 - Chengdu, China
Duration: 9 Jul 201011 Jul 2010

Publication series

NameProceedings - 2010 3rd IEEE International Conference on Computer Science and Information Technology, ICCSIT 2010
Volume9

Conference

Conference2010 3rd IEEE International Conference on Computer Science and Information Technology, ICCSIT 2010
Country/TerritoryChina
CityChengdu
Period9/07/1011/07/10

Keywords

  • Byte transfer layer(BTL)
  • Component
  • Framework
  • Heterogeneity
  • MCA
  • Module
  • Network extend method
  • Open MPI

Fingerprint

Dive into the research topics of 'A network extend method of OpenMPI'. Together they form a unique fingerprint.

Cite this