Twenty-Four Disks, Two Arrays, One Weekend Promise

The decode

RAID 5 survives one missing member per array, but not two. An array may continue operating with only one failed member; however, a second failure in the same array will take the array down. Regardless of what caused these failures, the recovery question is how many of the twenty-four disks read cleanly. A twelve-member RAID 5 is already at the extreme end of that design, stretching one disk's worth of parity protection across eleven neighbouring disks.

All the disks in an array are the same age, and that is not a comfort. Because all the disks in a RAID system are purchased together, from the same production batch, and have been performing the same task for the same amount of time, they are also typically at similar levels of wear. Therefore, if one disk fails due to wear, it is likely that the other disks are approaching their own failure point. This is precisely why the second disk so often fails during the stress of a rebuild.

A weekend deadline can only honestly be promised after imaging, never before. When all twenty-four disks image cleanly, completing the entire recovery within a weekend is realistic. However, when two members of one array are having difficulty producing readable data, extracting usable sectors from those two failing disks will set the speed of the recovery, and no promise made on Tuesday changes that.

Nothing is ever put back together on the original disks. Each member gets imaged first, and the arrays are rebuilt virtually from those images. Forcing a rebuild on the original hardware writes to every disk at once, which makes it the most common way a recoverable RAID case turns into an unrecoverable one.

Parity layout and member order are read from the disks, not assumed. The controller metadata, stripe size and rotation, as well as member order, are all determined by reading them off the disks. An XOR consistency check is then performed before a single file is extracted, to verify that there are no inconsistencies. Otherwise an incorrectly assembled array could give you corrupted files that look okay.

An honest quote cannot be just one number given before the disks are assessed. A single figure given before the members are assessed will either be priced for the worst-case scenario or gamble on the best case. A solid quote prices the imaging first, then fixes the recovery cost based on the actual condition of both arrays.

On the bench

Each member was catalogued, serial-matched to its slot on the shelf, and imaged individually under hardware write-blockers: healthy members directly, degraded members on DeepSpar Disk Imager with per-head control, and any member that behaved unstably through PC-3000 UDMA. Once every member had been imaged, both arrays were assembled virtually in UFS Explorer RAID, with the layout checked against the controller metadata and cross-checked in R-Studio Technician. They were then verified stripe by stripe using XOR consistency checks before the data was extracted onto new media.

The outcome

Twenty-four members were imaged individually, the two arrays assembled virtually from those images, verified with parity checks, and then extracted. The assessment is free, with one fixed written figure, plus VAT; where a drive has to be opened, 50% of parts and labour is payable upfront, with the balance only if we recover your data. The decode: the deadline belongs to the weakest disk. Image first, and the calendar becomes a promise you can keep rather than a risk you are taking with someone else's arrays.

Our case files are written up from genuine enquiries our lab has handled for customers across Sheffield and South Yorkshire, anonymised to protect client confidentiality. Each one describes the diagnostic and recovery approach our engineers apply to that fault, using the equipment listed.