基于消息传递的数据交错重分布负载平衡技术

Load balancing method with interleaving redistributed data based on message passing interface

数据重分布是实现消息传递环境下负载平衡的重要手段,提出了数据交错分布的模型问题及模型问题的并行计算模型,分析了模型问题在消息传递环境下的实现,讨论了性能和适用条件,给出了分析结果,讨论了通信与计算的时间重叠问题,将数据交错重分布负载平衡技术应用到非平衡刚性动力学方程组的并行计算中,获得了很好的负载平衡效果。

Data redistributing is the key technique to implement the load balancing in parallel programming based on message passing interface. A load balancing method with interleaving redistributed data suitable for message passing parallel computing environments is presented to solve the load imbalance problem arising from the parallel numerical simulation of non-balance dynamic equations to simulate non-balance phenomenon of radiation. At first, a load imbalance model is designed to describe the mam characteristic of the physical problems. For this model, the defined domain is uniformly differenced into grid cells, and every grid cell is assumed to process in different CPU time that can't be predicted by using the history information. Secondly, a load balancing method with interleaving redistributed data is presented. Next, an parallel model is given to analyse the performance of the new load balancing method. In the last, the new load balancing method is applied to the parallel numerical simulation of non-balance dynamic equations, and the performance is improved about 20%.