A performance bound analysis of multistage Combining networks using a probabilistic model

Byung Chang Kang, Gyungho Lee, Richard Kain

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

    2 Citations (Scopus)

    Abstract

    Combining networks have been suggested to improve the performance of shared memory MIMD computer systems, which can be seriously degraded when many processors access a shared variable at virtually the same time. Although it has been claimed that combining improves performance, the performance improvement has not been fully analyzed because of the complicated nature of combining networks. In this paper we introduce a probabilistic model to evaluate the performance of combining networks quantitatively. Using the probabilistic approach, we show a performance bound of multistage networks that use 2x2 combining switches. The bounds of performance improvement, in terms of network bandwidth and delay gain, are presented. We also discuss the effect of changing the combining queue size on the performance of combining networks.

    Original languageEnglish
    Title of host publicationICS 1991 - Proceedings of the 5th International Conference on Supercomputing
    EditorsEdward S. Davidson, Friedel Hossfield
    PublisherAssociation for Computing Machinery
    Pages448-457
    Number of pages10
    ISBN (Print)0897914341, 9780897914345
    DOIs
    Publication statusPublished - 1991 Jun 1
    Event5th International Conference on Supercomputing, ICS 1991 - Cologne, Germany
    Duration: 1991 Jun 171991 Jun 21

    Publication series

    NameProceedings of the International Conference on Supercomputing

    Other

    Other5th International Conference on Supercomputing, ICS 1991
    Country/TerritoryGermany
    CityCologne
    Period91/6/1791/6/21

    Bibliographical note

    Publisher Copyright:
    © 1991 ACM.

    ASJC Scopus subject areas

    • General Computer Science

    Fingerprint

    Dive into the research topics of 'A performance bound analysis of multistage Combining networks using a probabilistic model'. Together they form a unique fingerprint.

    Cite this