RAIDX: RAID without Striping

Each disk of traditional RAID is logically divided into stripe units, and stripe units at the same location on each disk form a stripe. Thus, RAID striping forces homogeneity on its disks. Now, consider a heterogeneous array of hard disks, solid state disks, and RAM storage devices, with various access speeds and sizes. If this array is organized as a RAID system, then larger disks have wasted space and faster disks are under utilized. This paper proposes RAIDX, a new organization for a heterogeneous array. RAIDX disks are divided into chunks, larger disks have more chunks. Chunks from one or more disks are grouped into bundles, and RAIDX bundles chunks of data across its disks. The heterogeneity of disks causes unbalanced load distribution with some under-utilized disks and some bottleneck disks. To balance load across disks, RAIDX moves most frequently accessed chunks to under-utilized, faster disks and least frequently used chunks to larger, slower disks. Chunk remapping is done at the RAIDX level and does not impact file system to storage addressing. Experiments comparing local and networked RAIDX against local RAID show that RAIDX has faster throughput than RAID when the array is composed of heterogeneous disks: local RAIDX is 2.5x faster than RAID, networked RAIDX is 1.6x faster than local RAID.