Based On Application And Optimization Of SMP Cluster Parallel Programming Model

With the development of science and technology, the demand for large-scale dataprocessing of every fields is increasing. And this has greatly promoted the development ofparallel computing. At the same time, the high performance computer is widely used. Amongthem, SMP cluster is the most popular computer system because it has high performance withlow price, good scalability and availability. For this reason SMP cluster become the mainstreamin the field of parallel computer architecture and a research focus in the field of highperformance computing. The SMP cluster and programming model suitable for the clusterarchitecture is the main research contents of this paper. And put forward the optimizationmeasures.Firstly, this paper introduces the research background and the development of parallelcomputing and parallel computer architecture. Then introduced several contemporary popularparallel computer systems, and analyzed the structure and characteristic of them. At the sametime, the current mainstream parallel programming model is introduced in this paper. Then thepaper searched and analyzed the key point in detail: SMP cluster and programming modelsuitable for the cluster architecture. The structure of the SMP cluster is introduced in detail inthis paper. The SMP cluster combines the advantages of SMP system and cluster system, thenode of SMP cluster is SMP and it has shared storage in nodes and distributed storage betweennodes, So we can use the shared memory programming model and distributed memoryprogramming model in the SMP cluster. OpenMP and MPI which represent the twoprogramming model is also the focus of this paper. In the light of the special structures of SMPcluster, given the hybrid programming model of MPI+OpenMP programming model. And giventhe structure, the realization mechanism and its advantages of hybrid programming model.Analysis the advantages and disadvantages of MPI+coarse-grained OpenMP and MPI+fine-grained OpenMP. At the same time puts forward some optimized measures to improve theperformance of hybrid programming model.The experimental part of this paper describes the hardware and software environment ofexperimental above all. And then introduced the method of performance evaluation. The mostimportant of experiment is achieving the PP algorithm parallel programming of N-body problem.And pure MPI and OpenMP+MPI parallel programming program are designed for them. In theend the mixed programming are optimized and the programs are achieved in the dawn ofTC5000fleet system. By the testing we can get the program execution time, and then find out thespeedup of program. The results proved that hybrid programming model can be further improvedby optimizing performance under certain circumstances. And we can also concluded that: MPI+OpenMP hybrid programming model is better than pure MPI programming model. Finally,the main contents in front of this paper are summarized and the work that can be carried out inthe near future is prospected.