Supercomputing pp 544-551 | Cite as

Communication and control costs of domain decomposition on loosely coupled multiprocessors

  • Luigi Brochard
Session 6A: Problem Mapping And Scheduling
Part of the Lecture Notes in Computer Science book series (LNCS, volume 297)


Communication and synchronization costs are a key problem in parallel computing. Studying an iterative method which requires only neighbor to neighbor communication, domain decomposition, we give different evaluations of speed-up depending on computation, communication and control costs. The effect of three different control costs for an iterative algorithm termination detection is studied. A linear control formulation is compared to experimental results on a loosely coupled array of eight processors. From the derived parameters, results for massively parallel systems are extrapolated.


Communication control domain decomposition iterative method lossely coupled multiprocessor parallel computing speed-up 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. [1]
    Andre F., Herman D., Verjus J.P. Synchronization of parallel programs, The MIT Press, 1985.Google Scholar
  2. [2]
    Axelrod T.S. Effects of synchronization bariers on multiprocessor performance. Parallel Computing, Vol 2, No 3, North-Holland, 1986, pp. 129–140.Google Scholar
  3. [3]
    Brochard L. Domain decomposition and relaxation methods, Parallel Algorithms and Architectures, M. Cosnard Ed., Elsevier Science Publishers B.V., North-Holland, 1986, pp. 61–72.Google Scholar
  4. [4]
    Brochard L. A shared memory MIMD computer: the Ultracomputer. La Recherche Aerospatiale, Paris, to appear.Google Scholar
  5. [5]
    Darema F. Applications environment for the IBM research parallel project (RP3). Proceedings of International Conference on Surpecomputing, Athens 1987, Lecture Notes in Computer Science, Springer-Verlag, New York, to appear.Google Scholar
  6. [6]
    Gannon D., Rosendale J.V. On the impact of communication complexity on the design of parallel numerical algorithms. ICASE Report 84-41, NASA Langsley, 1984.Google Scholar
  7. [7]
    Glowinski R., Dinh Q.V., Periaux J. Domain decomposition for elliptic problems, Finite Elements in Fluids, Vol 5. Gallagher Ed. Wiley & Sons, 1984, pp. 45–105.Google Scholar
  8. [8]
    Gottlieb A., Lubachevsky B.D., Rudolf L. Basic techniques for the efficient coordination of very large number of cooperating sequential processors. ACM Trans. on Prog. Lang. and Syst., Vol 5, No 2, 1983, pp. 164–189.Google Scholar
  9. [9]
    Rodrigue G., Sailor P. Inner/outer iterative methods and numerical Schwarz algorithms. Proc. of I.B.M. Conf. on Vector. and Parallel. Comput. for Scient. Comput., Rome, 1985.Google Scholar
  10. [10]
    Saad Y. Data communication in hypercubes. Research Report Yale Univ. DCS/RR-428, 1985.Google Scholar
  11. [11]
    Schwarz H. Gesammelte Mathematische Adhandlungen, Vol. 2, Berlin, Springer, 1890, pp. 133–134.Google Scholar
  12. [12]
    Widlund O. B. Iterative methods for elliptic problems on regions partitioned into substructures and the biharmonic Dirichlet problem. Proc. of Sixth Int. Symp. on Comp. Math. in Appl. Sci. and Eng., Glowinski and Lions Ed., 1984, pp. 33–45.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1988

Authors and Affiliations

  • Luigi Brochard
    • 1
  1. 1.Ecole Nationale des Ponts et Chaussées La CourtineNoisy le Grand CedexFrance

Personalised recommendations