Performance of a novel algorithm for perpendicular magnetic recording simulation

Abstract

The Grand Challenge problem presented by simulation of perpendicular magnetic recording requires an algorithm scaling close to linearly with the number of elements in the simulation cell. We derive an appropriate algorithm based on 2D FFT's of field and magnetization in the disk plane, combined with a novel treatment of field propagation in the z-direction. The near-linear scaling of the algorithm with number of elements is demonstrated by computer experiments. Although the computation contains a strong dataparallel element, in that an ensemble of 20 members is required in order to achieve statistics, some parallelization is desirable. Adequate parallelization is achieved on an SP machine. On an Intel cluster, speed-up saturation is observed, and is understood in terms of bandwidth limitations.