Accurately measuring MPI broadcasts in a computational grid Metadata

Metadata describes a digital item, providing (if known) such information as creator, publisher, contents, size, relationship to other resources, and more. Metadata may also contain "preservation" components that help us to maintain the integrity of digital files over time.

Title

  • Main Title Accurately measuring MPI broadcasts in a computational grid

Creator

  • Author: T, Karonis N
    Creator Type: Personal
  • Author: de Supinski, B R
    Creator Type: Personal

Contributor

  • Sponsor: United States. Department of Energy. Office of the Assistant Secretary for Defense Programs.
    Contributor Type: Organization
    Contributor Info: USDOE Office of Defense Programs (DP) (United States)

Publisher

  • Name: Lawrence Livermore National Laboratory
    Place of Publication: California
    Additional Info: Lawrence Livermore National Lab., CA (United States)

Date

  • Creation: 1999-05-06

Language

  • English

Description

  • Content Description: An MPI library's implementation of broadcast communication can significantly affect the performance of applications built with that library. In order to choose between similar implementations or to evaluate available libraries, accurate measurements of broadcast performance are required. As we demonstrate, existing methods for measuring broadcast performance are either inaccurate or inadequate. Fortunately, we have designed an accurate method for measuring broadcast performance, even in a challenging grid environment. Measuring broadcast performance is not easy. Simply sending one broadcast after another allows them to proceed through the network concurrently, thus resulting in inaccurate per broadcast timings. Existing methods either fail to eliminate this pipelining effect or eliminate it by introducing overheads that are as difficult to measure as the performance of the broadcast itself. This problem becomes even more challenging in grid environments. Latencies a long different links can vary significantly. Thus, an algorithm's performance is difficult to predict from it's communication pattern. Even when accurate pre-diction is possible, the pattern is often unknown. Our method introduces a measurable overhead to eliminate the pipelining effect, regardless of variations in link latencies. choose between different available implementations. Also, accurate and complete measurements could guide use of a given implementation to improve application performance. These choices will become even more important as grid-enabled MPI libraries [6, 7] become more common since bad choices are likely to cost significantly more in grid environments. In short, the distributed processing community needs accurate, succinct and complete measurements of collective communications performance. Since successive collective communications can often proceed concurrently, accurately measuring them is difficult. Some benchmarks use knowledge of the communication algorithm to predict the timing of events and, thus, eliminate concurrency between the collective communications that they measure. However, accurate event timing predictions are often impossible since network delays and local processing overheads are stochastic. Further, reasonable predictions are not possible if source code of the implementation is unavailable to the benchmark. We focus on measuring the performance of broadcast communication.
  • Physical Description: 96 Kilobytes pages

Subject

  • STI Subject Categories: 99 General And Miscellaneous//Mathematics, Computing, And Information Science
  • Keyword: Implementation
  • Keyword: Communications
  • Keyword: Algorithms
  • Keyword: Benchmarks
  • Keyword: Performance

Source

  • Conference: Eighth International Symposium on High-Performance Distributed Computing, Redondo Beach, CA (US), 08/03/1999--08/06/1999

Collection

  • Name: Office of Scientific & Technical Information Technical Reports
    Code: OSTI

Institution

  • Name: UNT Libraries Government Documents Department
    Code: UNTGD

Resource Type

  • Article

Format

  • Text

Identifier

  • Report No.: UCRL-JC-133177--Rev-1
  • Report No.: DP0101031
  • Grant Number: W-7405-ENG-48
  • Office of Scientific & Technical Information Report Number: 12133
  • Archival Resource Key: ark:/67531/metadc625016