Research And Implementation Clustered Parallel System Based On Linux And MPI

With the development of computer technology and its wide application, our dependency on the computer technology becomes more and more high. Computer's reliability and availability are becoming more and more important. Though a single computer's capability and reliability is becoming better and better, But a single computer can't complete some request still. At this time, it's necessary to organize several computers to work together. This technology is called cluster. The cluster technology is a technology of providing more availability and more reliability.Today in the IT field, the word cluster is not strange to us, especially in the the field in which server and high performance computing application was deeply accessible, the conception of cluster is very well-know. Since 80s, many kinds of cluster technology come forth in succession which rooted in the Digital VAX flatroof. The cluster technology becomes the mainstay of enterprise application and ISP computing and turns into the focus of researching about High performance computing and mass transaction processing because the cluster application is High usability and effective retractility.In this paper, we discusse structure and function of linux cluster in details, analyse and the research of cluster in existence and the program model of parallel computing. On the basis of deeply research and analysis framework and the technology of cluster middleware, we propose and design a new type of parallel computing system-based MPI and linux cluster. This paper analysises basic principles, ideas, composition and the main components' functions of the of parallel computing system-based MPI and linux cluster, and resolves some key technical problems about cluster middleware and parallel computing . In this parallel computing system , the Matrix multiple algorithm is realized and executed utilizing the MPICH. This system has a performance testing model for cluster network capability, and it can dynamically reflect the change features of capability parameter in thecluster. It has some good characteristics such as simplification, easy extensibility etc. This research and implementation works in the paper are a new research subject on parallel computing system based cluster . They shed lights on some new thoughts and plans for relative research.