RAID 0 Has No Spare Half

The decode

This is what makes RAID 0 so fast, and so dangerous: it splits every file across both drives. It breaks each file into chunks and puts alternate chunks on each disk, so it can read and write in parallel. That gives RAID 0 more speed, but there's zero redundancy. The data is never duplicated and there's no parity. So the array is weaker than a single drive. Both drives have to survive for any file to be whole.

When one disk dies, the whole volume goes down, not just that disk's share. Every file was striped alternately across both disks, so once one of them dies completely in a two-disk RAID 0 setup, you lose the whole volume. The good disk holds half of each file, but none of the files are complete. So "the Windows drive failed" doesn't mean you get back half the Windows data and keep all the Mac data safe. Both partitions are striped across both physical drives, and neither will work again until the bad disk has been imaged.

The restoration depends entirely upon the failed disk, which is the actual hard work. The good member images easily; the success of the whole process depends on getting an adequate image of the bad one. First, the type of failure is determined (electronic, firmware or mechanical), and the imaging method depends on that, because the reconstructed volume can only be as complete as the poorer of the two images. Everything else is reconstruction once that image exists.

All the steps to reassemble the volume are done virtually, never on the actual drives, and exactly to spec. Once both drives have been imaged, the striped volume is rebuilt from those images. Stripe size, block order and member order are all read from the image data and checked, since wrong values give corruption that looks fine. The Windows and Mac partitions are then copied out onto new media from the correctly assembled volume.

How complete the result is depends on how bad the failed drive is. If the failed member images fully, the whole stripe rebuilds and both partitions come back in full. If parts of it can't be read, the damage spreads across both partitions, because striping scatters it over the whole volume instead of keeping it in one place. You'd get holes in files all through both partitions, and we report those damaged files honestly, named from the rebuild.

RAID 0 will teach you this lesson eventually. The whole idea of RAID 0 is that redundancy is taken out on purpose. So it needs a backup as much as, if not more than, a single hard drive does, never less. Once your files are recovered, they go onto redundant backup media, and your RAID 0 array isn't rebuilt until there's a reliable backup behind it.

On the bench

Each of the two drives was imaged separately under hardware write-blockers: the good drive directly, while the bad drive was first evaluated on PC-3000 UDMA, then, being mechanical, opened on our head-swap bench for donor parts and imaged on DeepSpar Disk Imager with head mapping. Both images were then reassembled virtually into a complete stripe using UFS Explorer RAID, the stripe size and order were verified against the data, and the Windows and Mac partitions were extracted from the recreated volume and verified by hashing.

The outcome

We imaged both hard drives, repaired the failed drive mechanically, and then put together a virtual copy of the stripe, extracting both partitions along the way. The assessment costs nothing, and there's a single fixed quote for the work (plus VAT). Where we have to open up a drive, you pay 50% of the quote at the start, and only pay the rest if we recover your data. The decode: a striped array has no spare half. When one member fails, the whole striped volume goes down. So the real issue is making an accurate image of the failed drive. Reassemble it correctly and both partitions come back together.

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