We have presented a small but important change to the storage interface, known as range writes. By allowing the file system to express flexibility in the exact write location, the disk is free to make better decisions for write targets and thus improve performance.
We believe that the key element of range writes is their evolutionary nature; there is a clear path from the disk of today without range writes to the disk of tomorrow with them. This fact is crucial for established industries, where change is fraught with many complications, both practical and technical; for example, consider object-based drives, which have taken roughly a decade to begin to come to market [13].
Interestingly, the world of storage may be in the midst of a revolution as solid-state devices become more of a marketplace reality. Fortunately, we believe that range writes are still quite useful in this and other new environments. By letting the storage system take responsibility for low-level placement decisions, range writes enable high performance through device-specific optimizations. Further, range writes naturally support functionality such as wear-leveling, and thus may also help increase device lifetime while reducing internal complexity.
We believe there are numerous interesting future paths for range writes, as we have alluded to throughout the paper. The corollary operation, range reads, presents new challenges but may realize new benefits. Integration into RAID systems introduces intriguing problems as well; for example, parity-based schemes often assume a fixed offset placement of blocks within a stripe across drives. An elegant approach to adding range writes into RAIDs may well pave the way for acceptance of this technology into the higher end of the storage system arena.
Finding the right interface between two systems is always challenging. Too much change, and there will be no adoption; too little change, and there is no significant benefit. We believe range writes present a happy medium: a small interface change with large performance gains.