Research Of Performance Tuning Technology For High Performance Computing Cluster System

The cluster performance optimization technique is based on the method of common performance optimization, and it focus on the solving the problem of specific environment of the industry seismic processing application. According to the running status analysis of seismic jobs, we obtain the resource characteristics of various job types. And we also get the running time relationship of whole process flow in various jobs. Therefore the bottleneck of the seismic processing flow can be fund.The disk striping technique can do multi-disk concurrent read and write. It raises efficiency of 30% for the jobs on data buffer I / O.According to various job resource characteristics, the job schedule strategy is made to solve the problem of cluster nodes load balancing and contention. It improves the system throughput and raises the multi-tasking capability 45%.The characteristics of the various file systems are studied in this paper. Because seismic processing needs high-bandwidth for computer system, I study the cluster storage of key techniques. The distributed storage provides the high-bandwidth for seismic data processing. It solves the problem of bottlenecks integrated storage. While the cluster system is extended, the computing capability grows in linear.