Reliability Analysis of Declustered-Parity RAID 6 with Disk Scrubbing and Considering Irrecoverable Read Errors

We investigate the impact of Irrecoverable Read Errors (IREs) on Mean Time To Data Loss (MTTDL) of declustered-parity RAID 6 systems. By extending the analytic model to study the reliability of RAID 5 systems from Wu et. al. we obtain the MTTDL which mainly takes into account two types of data loss: data loss caused by three independent disk failures, and data loss due to a detected IRE during the rebuild after two disks failed. Furthermore we improve the analysis by also considering disk scrubbing to reduce the probability of IREs via periodically reading the data stored on a disk. The results of our numerical analysis show that IREs have a large effect on the MTTDL. The countermeasure is to increase the disk scrubbing rate. As an example, the MTTDL of a system where each disk is scrubbed everyday increases by a factor of at least 27 compared to that of a system with a scrubbing rate of once a year. In addition, declustered-parity RAID 6 system improves the reliability of standard non-declustered RAID 6 systems. For example, a declustered-parity RAID 6 system without disk scrubbing improves the MTTDLs by a factor at least 150 compared to that of a standard system where each disk is scrubbed everyday.