Seismic data in the field of oil and gas exploration is a typical kind of big data. The I/O storage system has become a major performance bottleneck of the whole system when pre-stack migration imaging is processed for seismic data using large scale parallel processing application software. Adopting the key technologies of the storage host cluster, dynamic storage multipath and parallel file system, this paper puts forward a storage framework model, analyzes the superiority, and does its deployment. The function and performance of this system is tested in typical application scenarios and the results verify the advantages of the system.