Wavepacket simulations in a message-passing cluster environment using a multi-agent-based loop scheduling

Abstract

This paper presents a computational strategy for improving the performance of wavepacket simulations using the quantum trajectory method (QTM). A multi-agent-based loop scheduling technique is utilized to dynamically load balance the parallel computation of the trajectories. The load imbalance introduced by the parallel runtime system and by the computationally-intensive moving weighted least squares (MWLS), the governing equations for the trajectories of the pseudoparticles, is reduced by the multi-agent-based loop scheduling at higher number of processors. Results of experiments on a multi-purpose message-passing cluster are presented to confirm that the multi-agent-based loop scheduling achieves better performance than with the current best adaptive loop scheduling technique when simulating a free particle at higher number of processors.