Issues in OR Parallelism

• Speculation (e.g., p in example).
  • To guarantee speedup: avoid speculative work.
    Too strong/difficult a restriction?

  • To guarantee no-slowdown:
    • Left-biased scheduling.
    • Instantaneous killing on cut.

• Granularity: avoid parallelizing work that is too small.

• Parallelization can be done:
  • Adding parallel/1 annotations to selected predicates (ECLPS)
  • Some systems attempt to do it automatically via the scheduler.

• Useful supporting techniques identified:
  • Visualization / trace analysis: Parsee [ECRC], VisAndOr and IDRA [UPM]
  • Program transformation to increase granularity. [ECRC]
  • Compile-time analysis to automatically introduce parallel annotations. [UPM]

• Quite successful systems built
  (ECLPS, SICSTUS/MUSE, Aurora, etc.)

• Significant speedups w.r.t. state-of-the-art Prolog/CLP systems can be obtained for search-based applications.