Parallel Sorting

• For small n the counting sort -- compute the final position of each element by n2 comparisons -- probably works well enough

  • A ZPL solution (based on the Problem Space Promotion idea) was given in Lecture #3

• For huge n -- greater than will fit in all memory -- a merging algorithm is probably best

  • Order local elements independently on each processor -- perfect parallelism
  • Merge the P lists
  • Probably constrained by disk speed

• Any algorithm using compare/exchange can use merge/split to handle larger n

Previous slide

Next slide

Back to first slide

View graphic version