Heterogeneous Beowulf Cluster-based Parallel Computing Model

Along with the technical development,the humanity has more and more big demand to the high performance computing.Heterogeneous computing can match appropriate architecture for different type computation,that reduces the accomplish time of computing makes the heterogeneous computing become the main research platform of parallel processing greatly. Heterogeneous computing have far-ranging foreground of application.Due to the upper ratio of capability and price,lower price and better scalability and other factors,cluster becomes hotspot research in the area of parallel computing.The factors of mature cluster system,depressed cost of hardware,high-speed advance of microprocessor capacity,mature LAN technology such as Ethernet and etc extend way for setting up cluster of parallel computing by microprocessor, namely,Beowulf system.We can make up from prime definition of Beowulf cluster to different deviser define went up Beowulf personally.The Beowulf cluster obtains flourishing development.The parallel computing model is a bridge connecting software and hardware,form pure computing model through abstract basal character from all kinds of parallel computers. Analyzing and designing algorithms which are carried out by hardware through compiled advanced language.But there is no all-purpose parallel model up to the present in despite of it is very important.The existent parallel computing models have different emphases,applicable environment,as well as heterogeneity,scalability,message communicating and other sides of applicable environment.Researching parallel computing model in the environment of Beowulf cluster has great meaning.In this thesis heterogeneous computing system,cluster system,Beowulf cluster and existing parallel computing models are analyzed,the HBSP parallel computing model is introduced with emphasis.Through to the study of HBSP system,the computation cost and communication cost in the barrier synchronization of HBSP system,heterogeneous Beowulf cluster is built,which manifests heterogeneous of system by the computation ability of node,and the nodes in the system have different CPU and memory capacity.Redefining HBSP system and its capability parameters,and introducing MPI into the HBSP model,as well as analyzing communication factors and partial computation of HBSP in the heterogeneous Beowulf cluster environment.The computation cost and communication cost in the barrier synchronization have been educed.Parallel algorithm of computing the value ofπis analyzed and simulated in the HBSP system based on heterogeneous Beowulf cluster with parallel software of MPI and program language of C++.The conclusion is worked out that the amended HBSP parallel computing model apply heterogeneous Beowulf cluster environment setting up in this thesis on a certain extent.At one time.the capability model is explored in this thesis.and the iso ratio of parallel overhead to computation is reviewed and amended in the foundation of experiment environment in this paper.We obtain the new extensible function that applies the heterogeneous Beowulf cluster.The amended iso ratio of parallel overhead to computation scalability model has good applicability in the heterogeneous Beowulf cluster environment by the experiment.