Abstract: A method and the associated problems for load balancing of CFD calculation on cluster of computers is described. When the domain decomposition method is adopted to solve the PDEs, the load on each processor of the cluster will be in proportion to the number of mesh cells in the sub-domain which is processed by that processor; therefore the equally partitioned sub-domains will make the load balanced. A load redistribution strategy is used to achieve dynamic load balance, for which a algorithm that can approximately equally partition multi-block mesh (which consists of structured mesh blocks, but can be combined in an unstructured way) into sub-domains is developed, and new data structure and program are made to ensure the correct linking and information communication at the inner boundaries between the sub-domains. Examples of partitioning some commonly used multi-block meshes with the algorithm are presented and a 3-dimensional CFD calculation is accomplished, The results show that the methods, algorithms and the programs are correct and can effectively balance the load, and that a quite good parallel efficiency is achieved.

Key words: parallel computing, computational fluid mechanics, domain decomposition, load balancing