Disk striping

Abstract

Just like parallel processing elements can substantially speed up computationally intensive tasks, concurrent transfer of data in and out of memory can speed up data intensive tasks. In this paper we study one general purpose facility for achieving parallel data motion: disk striping. A group of disks is striped if each data block is multiplexed across all the disks. Since each subblock is in a different device, input and output can proceed in parallel. With the help of an analytical model, we investigate the effect of striping on disk service times and its advantages and limitations in one of a set representative applications, file processing. We also explore several possible enhancements to striping: immediate reading, ordered blocks, and matched disks.