Statistical methods for automatic performance bottleneck detection in MPI based programs

Research output: Contribution to book/conference proceedings/anthology/reportConference contributionContributedpeer-review


With increasing number of processors the visualization of trace data for MPI communication primitives becomes harder due to many limitations. An automatic analysis of the recorded data should be able to identify some clues for the optimization of the source code. The approach presented with this paper is able to divide the communication time for each communication relationship into a necessary part and the overhead. We use statistical methods to define a time for each single communication event that we will accept as the 'usual' time for this event. Additional runtime will be considered as overhead. At the end of this article, the benefits from this method are demonstrated on the Sweep3D benchmark.


Original languageEnglish
Title of host publicationComputational Science - ICCS 2005: 5th International Conference
EditorsVS Sunderam, GD VanAlbada, PMA Sloot, JJ Dongarra
PublisherSpringer, Berlin [u. a.]
Number of pages8
ISBN (print)3-540-26032-3
Publication statusPublished - 2005

Publication series

SeriesLecture Notes in Computer Science, Volume 3514


TitleInternational Conference on Computational Science 2005
SubtitleAdvancing Science through Computation
Abbreviated titleICCS 2005
Conference number5
Duration22 - 25 May 2005
LocationEmory University
CountryUnited States of America

External IDs

Scopus 25144469347
WOS 000230414400041
ORCID /0000-0002-1686-8440/work/142240055


Library keywords