Research Of Key Technologies For Building Distributed System Based On Multi-core Processors

With the widespread application of multi-core technology, using multi-core computers in the design of distributed system has become a trend. In order to realize multi-level parallel and high-performance computing, this new architecture requires combining the advantages of multi-core processors and giving full play to the characteristics of distributed system.After the introduction of the basic concepts of the distributed system and multi-core processors, and the analysis of the research status of high-performance computing and building distributed systems based on multi-core nodes, a multi-layer model of the distributed system based on multi-core is proposed. The memory sub-model and the communication sub-model are specifically described. Combining the experience of programming practice, a hybrid mixed MPI with OpenMP is raised. The programming platform is applicable for the architecture of the distributed system based on multi-core nodes, and has the advantages of both the message passing and shared memory programming model. And on this basis, some specific problems, such as the implementation mechanism, control of the number of threads, optimized measures of the programming platform, are discussed. Using the method of MMTD, a replication algorithm which is suitable for the distributed system based on multi-core nodes is put forward. A new scalable model is also proposed. When the parallel system expands in its scale of machines and model of problems, the new scalable model can reflect the scalable characteristic of its performance more intuitively.In order to inspect and verify the correctness and effectiveness of the design of the distributed system based on multi-core nodes, a testing is done using a example of matrix multiplication. By comparing the performance of the distributed system under the different conditions of the serial, parallel and multi-level parallel, it confirms that the distributed system based on multi-core nodes has good computing performance, scalability, and fault tolerance.Through the example of the computing of matrix multiplication, this article also describes the process of multi-level parallel programming in detail, and does an experiment. The experimental results show that: compared with the traditional parallel computing, the distributed system based on multi-core nodes has better computing performance.