Parallelle beregninger med MPI i delt og distribuert minne

Abstract

Parallel Programming has been thrown upon us by the industry which can't deliver what we really want - ever faster CPUs. Instead they give us more computing cores at a steadily decreasing cost. Programmers are now forced to write parallel programs if we want to solve bigger problems. The purpose of this master thesis is to compare the two most popular programming platforms (cluster and multicore/shared memory) among the four available platforms (Cluster, multicore, GPU and cloud/internet). The comparison if not done solely on speedup/performance, but also on development time, ease of use, debugging time and size and type of problem solved. The master thesis will program a small selection of parallel algorithms on the four problems: Sorting, Matrix Multiplication, Delaunay Triangulation and the Mandelbrot set.

Parallel Programming has been thrown upon us by the industry which can't deliver what we really want - ever faster CPUs. Instead they give us more computing cores at a steadily decreasing cost. Programmers are now forced to write parallel programs if we want to solve bigger problems. The purpose of this master thesis is to compare the two most popular programming platforms (cluster and multicore/shared memory) among the four available platforms (Cluster, multicore, GPU and cloud/internet). The comparison if not done solely on speedup/performance, but also on development time, ease of use, debugging time and size and type of problem solved. The master thesis will program a small selection of parallel algorithms on the four problems: Sorting, Matrix Multiplication, Delaunay Triangulation and the Mandelbrot set.