Advertisement

Analysis of multigrid methods for non-shared memory systems by a simple performance model

  • O. Kolp
  • H. Mierendorff
  • W. Seidl
Namertal Algorithms (Session 1.2)
Part of the Lecture Notes in Computer Science book series (LNCS, volume 237)

Abstract

The system behaviour of parallel processors normally depends in a complex way on many parameters. We investigate the quality of a linear model for the transport in non-shared memory systems by the example of a multigrid method. To this purpose, this method was implemented on the processor kit DIRMU-25 for measuring the required computing time. On the other hand, we have developed a simple abstract process model describing the computational work by one parameter and the transport work by two parameters. The simulation of the abstract model is compared with the implementation at the measuring points. In this way, we obtain information about the quality of the model and the interpretation of the parameters used. The results of this comparison enable us to forecast system performance for large systems by simulation. Though not being optimal for the DIRMU-25 system, the used algorithm is sufficiently complex to illustrate, in particular, the transport characteristics of the system.

Keywords

Transport Cost Multigrid Method Quadratic Domain Multigrid Algorithm Relaxation Operator 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

References

  1. [CS85]
    T. F. Chan and Y. Saad: Multigrid Algorithms on the Hypercube Multiprocessor; Yale University, report no. YALEU/DCS/RR-368, 1985.Google Scholar
  2. [CS83]
    T. F. Chan and r. Schreiber: Parallel Networks for Multi-Grid Algorithms: Architecture and Complexity; Yale University, report no. YALEU/DCS/RR-262, 1983.Google Scholar
  3. [GK84]
    B. Görg and D. Kolp: Parallele Rechnerarchitekturen für Mehrgitteralgorithmen; in Trottenberg and Wypior (eds.): Rechnerarchitekturen für die numerische Simulation auf der Basis superschneller Lösungsverfahren 11, GMD-Studien no. 102, GMD, D-5205 St. Augustin, 1985.Google Scholar
  4. [GH84]
    L. Geus, W. Henning, W. Seidl and J. Volkert: MG00-Implementierungen auf EGPA-Mehrprozessorsystemen; GMD-Studien no. 102, GMD, D-5205 St. Augustin, 1985.Google Scholar
  5. [HM85a]
    W. Handler, E. Maehle and K. Wirl: DIRMU Multiprocessor Configurations; Proc. Int. Conf. on Parallel Processing, University Park, Pennsylvania, aug. 1985Google Scholar
  6. [HM85b]
    W. Handler, E. Maehle and K. Wirl: The DIRMU-Testbed for High-performance Multiprocessor Configurations; Proc. 1st Int. Conf. on Supercomputing Systems, St. Petersburg, Florida, dec. 1985Google Scholar
  7. [KM84]
    O. Kolp and H. Mierendorff: Systemunabhängige Organisation von Mehrgitterverfahren auf Parallelrechnern; in Ehrich (ed.): GI-14. Jahrestagung, Braunschweig, Okt. 1984, Proc.; Informatik-Fachberichte 88, Springer-Verlag.Google Scholar
  8. [KM85a]
    O. Kolp and H. Mierendorff: Efficient Multigrid Algorithms for Locally Constrained Parallel Systems; Proc. 2nd Copper Mountain Conf. on Multigrid Methods, march 31–april 3, 1985, Copper Mountain, Colorado; subm. to AMC.Google Scholar
  9. [KM85b]
    O. Kolp and H. Mierendorff: Bus Coupled Systems for Multigrid Algorithms; Proc. 2nd European Conf. on Multigrid Methods, Köln, oct. 1985; to appear.Google Scholar
  10. [ST82]
    K. Stüben and U. Trottenberg: Multigrid Methods: Fundamental Algorithms, Model Problem Analysis and Applications; in Hackbusch and Trottenberg (eds.): Multigrid Methods, Proc. of the Conf. in Köln-Porz, nov. 1981; Lecture Notes in Mathematics, Springer, Berlin 1982.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1986

Authors and Affiliations

  • O. Kolp
    • 1
  • H. Mierendorff
    • 1
  • W. Seidl
    • 2
  1. 1.Gesellschaft fuer Mathematik und Datenverarbeitung mbH Schloss BirlinghovenF. R. Germany
  2. 2.Institut für Mathematische Maschinen und DatenverabeitungUniversität Erlangen-NürnbergErlangenF. R. Germany

Personalised recommendations