Overcoming Fear Of Wasted Effort on MPP

• Worst-case scenario: MPP market collapses
  • Working on parallel programming reveals concurrency
    • Helps on PVP and SMP architectures

  • Working on MPP's with cache+superscalar nodes reveals caching strategies
    • Helps on workstation and SMP architectures

• Unpleasant scenario: My favorite MPP manufacturer collapses
  • Portable methods will port to survivor machines
    • If manufacturers ignore demands to support portable programming systems (MPI, HPF), they deserve to die.

  • Nonportable methods will not port to survivor machines
    • If the scientist-programmer "throws his body at the machine"*1 to program it, he gets hurt twice:
      • bouncing off

      • hitting the ground

  ---

  Scientific Computing: Where does MPP Fit In? - Lecture given 6/14/95
  [Next] [Previous] [Top]

  Generated with CERN WebMaker