Running MPI jobs with WebCom

Research output: Chapter in Book/Report/Conference proceedingsChapterpeer-review

Abstract

A "typical" MPI job consists of 4 components: an MPI binary; an optional machines file; a set of optional input files and a parameterised start script. Of these, the machines file and the starting script are most prone to change from run to run on the same parallel architecture or on a different architecture. Moving an MPI job to a different architecture will precipitate the greatest changes in these and possibly the other components. The work associated with making and managing these changes is tedious and error-prone, especially if the user-base is large, andlor the underlying system is dynamic in size or is heterogeneous. This problem can be addressed by employing an abstract machine to separate the MPI launch process from the underlying parallel architecture. In this role the WebCom abstract machine can readily be used to generate a parallel architecture from component machines, to dynamically create a machines file and startup script and, if necessary, to stage the binary and input files. Change management is also addressed by employing WebCom to manage user accounts, shared file space, data storage and on-going run-time costs.

Original languageEnglish
Title of host publicationProceedings of the 2005 International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA'05
PublisherCSREA Press
Pages946-949
Number of pages4
ISBN (Print)9781932415605
Publication statusPublished - 2005
Event2005 International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA'05 - Las Vegas, NV, United States
Duration: 27 Jun 200530 Jun 2005

Publication series

NameProceedings of the 2005 International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA'05
Volume3

Conference

Conference2005 International Conference on Parallel and Distributed Processing Techniques and Applications, PDPTA'05
Country/TerritoryUnited States
CityLas Vegas, NV
Period27/06/0530/06/05

Keywords

  • Cluster
  • Condensed graphs
  • Mpi

Fingerprint

Dive into the research topics of 'Running MPI jobs with WebCom'. Together they form a unique fingerprint.

Cite this