Data integrity analysis of disk array systems with analytic modeling of coverage

Detailed dependability models of various disk array organizations are developed taking into account both the hard disk failures and transient errors. Various error and failure modes of individual disks and the disk array are identified. A small proportion of transient errors account for uncovered failures in a disk array. This leads to analytic computation of probability of disk failures based on several factors including byte error rate of a disk and ECC code etc. Traditionally used measures like MTTDL and data availability remain virtually unchanged with change in mean recovery time. New dependability measures such as degraded capacity time are considered to bring out the effect of mean recovery time on dependability of disk arrays. Our analysis also reveals that mirrored disk organization has higher MTTDL than other disk array organizations if the only failure mode considered is data loss while catastrophic errors are ignored. However, if catastrophic errors are taken into account, then RAID-3,4,5 organizations have higher data-integrity than other disk array schemes. We also develop models that take into account the reliability of support hardware components and different placement schemes for arranging support hardware such as power supply. Our analysis reveals that RAID-1 benefits the most from orthogonal placement of support hardware.