基于多块结构网格大规模并行计算的负载均衡设计及实现

结构网格具有网格生成速度快、质量高、数据结构相对简单、较适用于流体表面应力集中的运算等优点。在大规模 CFD（Computational Fluid Dynamics）并行计算中，需要将网格区域划分为多块网格，而多块网格之间的数据通信会制约并行计算能力的提高，因此对结构网格的负载平衡优化是提高并行计算能力的重点。本文提出了一个采用多层次 k-way 多约束条件图剖分算法来改进负载平衡的方案，并对 M6 翼型和 CRM 模型的多种规模进行了实际计算，结果证明多层次 k-way图剖分算法能够有效地优化负载平衡，在此基础上得到了最优节点间的计算负载平衡和通信负载平衡，最终达到了理想的并行效率。

Load Balance Design of Large Scale Parallel Computing Based on Multi-Block Structured Grid and Its Implementation

Structured grid has the advantages of fast speed, high quality, simple data structures, and being suitable for fluid and surface stress concentration and other aspects of the CFD calculations. In large scale parallel computing, the grid area will divided into multi-block, however, the data communications between the blocks will restrict the computing efficiency and the parallel computing capability. Therefore, load balancing optimization of the structured grid is the key to improve the ability of parallel computing. This paper presents a multi-level k-way split map load-balancing algorithm to improve it. Practical calculations on F6 airfoil and CRM scale model were conducted, and the results showed that this approach can efficiently optimize the load balance, a better computing load balancing and traffic load balancing between nodes have been achieved, and so as a high parallel efficiency finally.