点对点通信原语并行转换方法仿真研究

针对传统点对点通信原语并行转换方法无法集中分析内部数据结构,导致整体转换效果较差的问题,提出一种MPI程序下点对点通信原语并行转换方法。分析当前原语代码数据结构,完成对应结构化操作,基于并行解码的上行数据交换数据,根据数据理论分析获取数据节点冲突概率,引入高密度MDSCAN聚类算法实现符号的数据簇分类,利用Omega数据库的运算通信模式转换原语,实现通信原语转换为原语数据集。实验结果表明,研究方法的原语数据集抗压比和数据贴合度明显提高,数据显著性更好,转换效果更理想。

Simulation Study on Parallel Conversion Method of Point-to-Point Communication Primitives

Traditional method of parallel conversion for point-to-point communication primitives is unable to analyze the internal data structure centrally, leading to the poor overall conversion effect. In this paper, a parallel conversion method for point-to-point communication primitives based on MPI program was presented. At first, the current primitive code data structure was analyzed and the corresponding structured operation was finished. Based on the parallel decoding, the uplink data exchanged the data. According to the theoretical analysis, the conflict probability of data node was calculated. After that, the high-density MDSCAN clustering algorithm was used to realize the classification of symbol data clusters. Finally, the operation communication mode of Omega data base was used to convert the primitives, and thus to realize the conversion of primitive data set. Simulation results show that the proposed method can significantly improve the compression resistance and data fitting degree of primitive data set. Meanwhile, the data significance and the conversion effect are better.