I don't really understand why this is 16x more tasks to be done when comparing to the sequential approach. Could someone elaborate on the detail of the sequential approach?
jkorn
@cwchang, the sequential approach would be to just go through every particle, see which cell it is within, and append the particle to cell_lists[c]. Now, to parallelize across 16 tasks (one per cell), in each task you are still going through the entire list of particles. So while you are parallelizing across 16 tasks, you are going through the entire particle list 16 times overall (instead of just once with the sequential algorithm), giving yourself 16x the work. 16x work/16 tasks = 1 so you basically get no speedup from this.
I don't really understand why this is 16x more tasks to be done when comparing to the sequential approach. Could someone elaborate on the detail of the sequential approach?
@cwchang, the sequential approach would be to just go through every particle, see which cell it is within, and append the particle to cell_lists[c]. Now, to parallelize across 16 tasks (one per cell), in each task you are still going through the entire list of particles. So while you are parallelizing across 16 tasks, you are going through the entire particle list 16 times overall (instead of just once with the sequential algorithm), giving yourself 16x the work. 16x work/16 tasks = 1 so you basically get no speedup from this.