The Single System Imagine Cluster System

In recent years, the high performance computing (HPC) attracts more and more attention in the world. The application of HPC in the science research and military affairs gets huge success. The cosmically parallel computing based on the HPC also gets many productions. With the development of research and application of HPC, the cluster gradually becomes the developing trend because the cluster has the good expansibility and cheaper price. The aim of the cluster is to make all nodes cooperate with each other, and the SSI which aiming at banding the correlative technology of cluster and achieve the really parallel computing is the hot spot of cluster research.On the base of the summarizing and analyzing of the status and development about the parallel computing currently, this paper introduces the various aspects of the SSI project, analyses the shortages of the LVS researching, and then describes the technique of implementing the group dispatcher which using duplicated MAC and IP address method in the4stack of TCP/IP. Moreover, we solve the shortage of the current technique through combining the load balance and NAS techniques.By using the group dispatcher, we solve the problem of the whole cluster down brought by the single centralized dispatcher's down. At the same time, it also solves the bottleneck problem of the performance of the cluster. And it will distribute the access to the different node and get the high utilization of system resource according the corresponding load balance algorithm. Also we adapt NAS technology to access the data in the backend node. The result of experiment shows that we can distribute the requests and access the data efficiently and can also improve the performance of the cluster by using the method in the condition of accessing cluster server, especially in the case that the accessing mount is very huge.