Parallelizing the Spot Model
Chris H. Rycroft, term project for 18.337 Parallel Computing
Abstract: The recently proposed spot model provides a method for the rapid simulation of flow in dense amorphous packings, and this has been implemented as a serial algorithm for the case of granular drainage from a silo. Since the spot model is based on a local microscopic interaction, it is hoped that it can be efficiently converted into an algorithm for a parallel computer. Here we analyze two possible parallelization schemes, one based on a master/slave architecture, and another based on distributed algorithm. The times of the two algorithms have been studied, and while far from optimal, significant improvements can be seen over the corresponding serial code. Further methods of improving the efficiency of a parallel spot algorithm are proposed.
Download: PDF (5.8Mb)