Tunable Sparing of Disks in a Cloud Data Center

Disk Drive failure tolerance and optimum utilization of disks for storage are two of the most important challenges in the Cloud based data center. Addressing both these problems simultaneously can be tricky. To make the system failure tolerant, more storage needs to be added; this results in redundancy of storage. Providing a safety net for the disk is important to make sure that the disk can still operate. Therefore, we propose a novel drive sparing mechanism which not only provides failure tolerance by maintaining the optimal number of spare drives in the storage system, but also deals with over-provisioned spare. The method first identifies the critical disks that may fail in the RAID group based on the Venn-ABERS framework which helps us to know about the conditional probability of the disk failure, and then implements a variable sparing mechanism where a few disks act as dual-purpose disks. We demonstrate variable sparing using two cases, first when minimum spare pool is sufficient, and second, when minimum spare pool is not sufficient for failure tolerance. By proactively addressing potential failure points, the drives are able to extend their lifespan and enhance reliability when the variable sparing is utilized. In this paper, we explain the mechanism of variable sparing of disks and share results of an evaluation experiment using an open-source dataset by utilizing our prototype system.