MPI on Hybrid Systems

MPI + OpenMP and other models on clusters of SMP nodes

14.01.02


Start Lecture by clicking here


Content

Author: Rolf Rabenseifner

  1. MPI on Hybrid Systems MPI + OpenMP and other models on clusters of SMP nodes
  2.     Outline
  3.     The Programming Models
  4.     Hybrid Systems
  5.     Programming Models on Hybrid Systems
  6.     Questions
  7.     Why MPI + OpenMP / Automatic SMP-parallelization ? (2)
  8.     MPI + OpenMP is really used!
  9.     Questions
  10.     How can I program these programming models (1)
  11. How can I program these programming models (2)
  12.     Hybrid Programming Technique (1)
  13.     Hybrid Programming Technique (2)
  14.     But results may surprise!
  15.     The heat example — MPI+OpenMP
  16.     The heat example — MPI+Compas (Hitachi)
  17.     Benchmarking example
  18.     MPI rules with OpenMP / Automatic SMP-parallelization (1)
  19.     MPI rules with OpenMP / Automatic SMP-parallelization (2)
  20.     MPI rules with OpenMP / Automatic SMP-parallelization (3)
  21.     Programming method / „provided“
  22.     Rules with MPI+OpenMP / Compas on Hitachi SR8000
  23.     (H1) Calling MPI only outside of parallel regions
  24.     (H1) Calling MPI only outside of parallel regions – example
  25.     (H2) Calling MPI also inside of OMP MASTER
  26.     (H2) … the barrier is necessary – example with MPI_Recv
  27.     (H2) … the barrier is necessary – example with MPI_Send
  28.     (H3) Calling MPI also inside of OMP SINGLE
  29.     MPI + OpenMP on Hitachi SR8000
  30.     Batchjob-Script Recommendation on Hitachi SR8000
  31.     Questions
  32. Which model is the best?
  33.     Advantages (+) and Disadvantages (–), I.
  34.     Advantages (+) and Disadvantages (–), II.
  35.     Advantages (+) and Disadvantages (–), III.
  36.     Advantages and Disadvantages, Summary
  37.     Summary

Back to the Parallel Programming Workshop Overview