Particle simulation using serial, GPU anduddistributed approaches

摘要

In computer engineering, simulation is a popular and reasonable method to studyudscientific problems, which evaluates the motion of different objects in various sizes ofudsimulation spaces. In order to achieve better performance, different approaches willudbe applied. Nowadays, both GPU and cluster show great power in parallel computing.udIn this thesis, a particle simulation is formulated by following the motion of interactingudparticles as they move in some constrained space, colliding with each otherudand the walls. We compare three solutions to this problem: i) using traditional (serial)udcomputing, ii) using general purpose computing on a graphics processing cardud(GPGPU), and iii) using a distributed cluster architecture and the message passingudinterface (MPI). Based on the experimental data gathered from the tests, the performanceudof the algorithms is analyzed to show how the speedup varies across differentudarchitectures and with the number of compute cores used.