Exploiting parallelization in spatial statistics: an applied survey using R

Abstract

Computing tasks may be parallelized top-down by splitting into per-node chunks when the tasks permit this kind of division, and particularly when there is little or no need for communication between the nodes. Another approach is to parallelize bottom-up, by the substitution of multi-threaded low-level functions for single-threaded ones in otherwise unchanged user-level functions. This survey examines the timings of typical spatial data analysis tasks across a range of data sizes and hardware under different combinations of these two approaches. Conclusions are drawn concerning choices of alternatives for parallelization, and attention is drawn to factors conditioning those choices.