A load balancing method based on dynamic time step feedback

Abstract

In dynamic environments, dynamic load balancing plays an important role in improving the performance of large-scale parallel computer systems. It can reduce the response time of the calculations and maximize the resources utilization of system. To improve the parallel efficiency in dynamic systems, a dynamic load balancing method with a prediction and dynamic time step feedback was developed. The feedback is varied dynamically by using a time step and processing capability prediction algorithm combined with a dynamic programming algorithm for task redistribution. Apply this method in ab initio molecular dynamics simulation, this algorithm showed significantly better adaptability and scalability compared to one that employs a static time step feedback.