A Clean-Slate Look at Disk Scrubbing

A number of techniques have been proposed to reduce the risk of data loss in hard-drives, from redundant disks (e.g., RAID systems) to error coding within individual drives. Disk scrubbing is a background process that reads disks during idle periods to detect irremediable read errors in infrequently accessed sectors. Timely detection of such latent sector errors (LSEs) is important to reduce data loss. In this paper, we take a clean-slate look at disk scrubbing. We present the first formal definition in the literature of a scrubbing algorithm, and translate recent empirical results on LSE distributions into new scrubbing principles. We introduce a new simulation model for LSE incidence in disks that allows us to optimize our proposed scrubbing techniques and demonstrate the significant benefits of intelligent scrubbing to drive reliability. We show how optimal scrubbing strategies depend on disk characteristics (e.g., the BER rate), as well as disk workloads.