Research Of Key Technology Of Parallel File System Based On High Performance Cluster Computing

This thesis presents an architecture of platform based on parallel file system forthe seismic data processing in field of oil and gas, it eliminates the bottleneck of sharednetwork storage system and prompts the efficiency of massive data processing. Thereare a large amount of concurrent read and wrote in process of seismic dataprocessing, but the limits of bandwidth of network, performance of file system lead tothe I/O wait and lower efficiency of cluster computing, and makes the collaborativework of geological exploration impossibleThe thesis analyzes the feature of parallel computing and data layout, popularnetwork storage technology in seismic data processing, discusses the key factors ofreducing the efficiency of high performance computing. By the thorough discussion ofLustre, a parallel file system,and its applicable scenarios, it is very essential for gettingthe high performance of cluster system to deploy the Lustre file system in storagesystem and applications.It is key to build a set of high performance cluster computing configuration, sothe test of aggregate I/O bandwidth, efficiency of the application of seismic dataprocessing is carried out, by analysis of the result, it proves the solution is reasonable,the promotion and optimization of the solution is possible in future.