一个面向数据密集型网格的通信模型

在由多计算机集群构成的数据密集型网格环境下,为了解决通信峰值阶段的通 信冲突问题,提出了一个多级分组通信模型MGCM(Multi-level Grouping Communica tion Model)).给出了由多机群组成的数据密集型网格数据节点DGDN(Data-intensive Grid Data Node)和完全图通信等形式化定义.利用主动队列、被动队列机制,研究了单个机群内的分组通信算法GCA(Grouping Communication Algorithm)以及双机群内的计算机之间的完全图通信问题GCAD(Grouping Communication Algorithm in Double computer clusters). 基于网络距离机制、GCA和GCAD,设计了DGDN上的多级分组通信模型.理论分析和实践表明 ,该模型有效地解决了数据密集型网格环境下通信峰值所造成的并行计算效率低下的问题. 该模型适合于基于网格的数据密集型并行计算.

Communication model for data-intensive grid

For resolving the conflict problem of the peak communication in data-intensive grid that composed of CC(computer clusters), a MGCM(multi-level grouping communication model) was discussed. A series of formal definitions, such as the DGDN(data-intensive grid data node) and the complete graphics communication were given. Using the mechanism of the active queue and passive queue, a GCA(grouping communication algorithm) in single computer cluster and a GCAD(grouping communication algorithm in double computer clusters) were studied. Based on network distance, GCA and GCAD,a MGCM of DGDN were designed. The analysis and experiment results show that MGCM effectively resolve the low efficiency during the peak communication in grid. It can be fit for the data-intensive parallel computing based on grid.