An Improvement Of Parallel Computational Model Based NHBL And It's Performance Evaluating

The kernel of this thesis is parallel computing models, as an important research area of parallel computing, which provide a directive theoretical interface and model framework for analysis and designing of both parallel algorithms and systems architectures. In the realm of sequential computing Von Neumann machine has successfully provided an unifying and universal model of computation. In the realm of parallel computing, however, there has been no similar success. So I expand this research based on parallel computing model.First the NHBL model (Nondedicated Heterogeneous Barrier LogGP model) has been introduced, and it is based on LogGP model . But LogGP doesn't be considered nondedication and heterogeneity between nodes, so it can't be used to describe the parallel computing time accurately. NHBL which based on LogGP model considers these, but it doesn't reflect waste to communication between nodes. Based this, a new model NHBCL(Nondedicated Heterogeneous Barrier Congestion LogGP model) has been put forward. This model uses the congestion exponent to describe the net congestion.First NHBL and NHBCL model have been described and defined their computation and communication cost models in details , then I forecast the parallel computing value ofπ. Then under the MPI environment , I get the true value ofπvia experiment .At last, I have a comparison between the true value and analyzing value ofπ. The experimental results show the NHBCL model commendably depicts the features of the parallel computing based on cluster, and it is practical and correct for the parallel computing based on cluster.