A New Load Balancing Method Based on Dynamic Cluster Construction for Solution-Adaptive Finite Element Graphs on Distributed Memory Multicomputers

Abstract

To solve the load imbalance problem of a solution-adaptive finite element application program on a distributed memory multicomputer, the load of s refined finite element graph can be redistributed, based on the current load of each processor. For this purpose a load-balancing algorithm can be applied to balance the computational load of each processor. In this paper, a distributed methods for load balancing is proposed, which is based on the global load balancing information and current load distribution of the system. A simulation model has been developed to computre the performance of the proposed methods with previously stated methods like Maximum cost spanning tree Load-Balancing (MCSTLB) method, binary tree load balancing (BTLB) method and condensed binary tree load balancing (CBTLB) method. Two criteria, the execution time and the number of process migration required by different load balancing methods have been used for performance evaluation. The experimental result shows that the execution time and the number of process migration required by the proposed method is better than that of existing methods.