With the development of high performance center build on server cluster, the speed of CPU and Memory have gained a big leap in massive data processing, but the unmatched I/O speed is still a bottleneck of whole system. This paper digs into streaming media system’s specific characters deeply in order to find out a more efficient, balanced way of file storage distribution, increasing degree of parallelism while reducing service time delay. First, give an evaluation of each media file based on target features, then distribute all files onto different disks through storage scheduler, finally there is a fine grit division to guarantee storage inside one disk is highly optimized. The simulation results verify the validity of the algorithm, this strategy of storage allocation not only improves data organization among disks, but also works well in one disk, which brings a better concurrent service and user experience.