Protecting Single Shingled Write Drives Against Latent Sector Failures

No ratings

Presented at Systor2018Haifa 2018 by

Thomas Schwarz

The recently introduced shingled write technology is necessary for reaching drive capacities of 10 TB, but no longer allows blocks to be updated in place. This inconvenience however simplifies the use of intra-disk redundancy to protect disks against latent sector failures, necessary when the disk drive does not form part of a larger ensemble. Latent disk sector errors usually befall single disk blocks and are only discovered when repeated attempts to read the block have failed. It is known that they afflict a sizable portion of modern disks. Because of the nature of shingled writing, content is written to disks in large contiguous bands. Redundant information can be calculated over large segments of each bandand embedded as parity blocks into the segment. We investigate strategies for generating redundant information, evaluate various codes using distributions of latent sector locations obtained by Schroeder and colleagues, and propose a simple pyramid code that combines easy creation and easy recovery of unreadable blocks with low storage overhead and good protection.