The I/O Problems Research In The Parallel Computing

With the development and progress of high-performance parallel systems,the application of parallel computing becomes important more and more, the requirement of improving the parallel computing's performance is increasing, at the same time,the I/O devices do not match the the growth rate of CPU's speed, it becomes the bottleneck to increasing the parallel computing performance,so the study is particularly important to the I/O problem of parallel computing.To the I/O problem research,the scholars of domestic and foreign have been the achievements, there is little result of studying the I/O problem by using MPI(Message Passing Interface) standard,this paper is a special subject research to the I/O problems in the parallel computing by adopting MPI as the programming environment and using two clusters as the experimental platform.In this paper, the I/O problems in parallel computing were analyzed. The introduction was made to parallel computing and cluster, the similarities and differences were clear between cluster one and cluster (?)o by comparing the two clusters of using in the experiment. Two basic(?)ays of parallel computing were described through experiments, experimental results show, the parallel I/O may achieve the better performance.The MPI-IO parallel I/O ways of MPI standard were described further in this paper, the description and implementation was done to the MPI-IO parallel I/O ways flexibility in the use of datatypes, the analysis of experiment was done to the MPI-IO parallel I/O ways specificity in the application of the collective I/O methods, the analysis and results of experiment show that solving the I/O bottlenecks problem may be achieved by the use of the MPI-IO parallel I/O ways flexibility and handling to the special application of the MPI-IO parallel I/O ways.