(80f) An Approach For The Parallelization Of The Discrete Element Code

Dubey, A. - Presenter, Rutgers University
Tomassone, M. S. - Presenter, Rutgers University
Muzzio, F. J. - Presenter, Rutgers University

A multi-stage approach for parallelization of a Discrete Element Methid (DEM) code has been developed. This algorithm consists of three stages with the first stage as the one that will reduce the simulation time the most. The multiple processor cluster has been divided among one lead processor (parent) and the remaining child processors which can number from four to over a hundred, depending on the availability. The main processor manages the neighbor lists of all particles and updates it periodically. It also distributes the particles equally among the child processors which do computations on only the particles that they are assigned. The message passing interface (MPI) algorithm is used to perform these scatter (dividing of particles) and gather (collecting data from the child processors). In stage 1, the parallelization involves only the distribution of particles among the processors which perform the dynamics calculations on the particles assigned to them serially. The dynamics calculations which involve calculation of parameters such as new particle position, force, velocity, temperature, humidity etc. are not parallelized. In stage two, these calculations are also parallelized to enhance the overall performance of the code. In the third stage, the handling and updating of neighbor list is performed by a number of processors instead of one main processor, hence further reducing computational times.