Tolerance, failure, geometry
Every level and layout, and what it takes to put each one back. The level is the promise. The layout is the arithmetic.
A RAID level is a promise about how many drives can go before the data does, and a layout is the arithmetic that keeps it. When the promise is broken, by a second failure during a rebuild, a controller that lost its configuration, a rebuild onto the wrong drive or a command typed at three in the morning, the arithmetic is what the bench works from: which drive held which stripe, where the parity went, how big the pieces were, and where the first block sat. Pick your level below. Each page says what it tolerates, what fails it, and what has to be determined to put it back together. A set of two to four members is £500 + VAT upwards after the free look, fixed in writing; larger sets, 15K SAS sets, storage shelves and enterprise storage start at £1,250 + VAT, and on most jobs no data means no bill.
Rather talk it through? An engineer answers the bench line
0800 6890668
The standard levels, or the software and vendor layouts.
Rather begin with what it says →The standard levels
RAID 5One parity block per stripe, rotating. Order, stripe size, rotation direction and symmetry; the stale member; the write hole→RAID 6P and Q. Two failures tolerated; Q must be checked to prove the geometry; the third failure→RAID 10Mirror pairs striped. Which drives are pairs, which copy is newest, near and far layouts on mdadm→RAID 1Two copies. The newest one, the offset the controller left, and the stale mirror rebuilt over the good one→RAID 0Nothing to check against. Order and stripe size from the file system's own structures and entropy→RAID 50 and RAID 60Spans of 5 or 6, striped. Per-span geometry, span order and the span stripe; shelves from £1,250 + VAT→Software and vendor layouts
RAID-Z1, Z2 and Z3ZFS. Variable stripe width, per-block parity, four labels per device and an uberblock ring; zpool import -F stays off the originals→SHR and Synology RAID F1mdadm arrays across partitions of mixed sizes, joined by LVM, with a RAID 1 system partition under it all→Storage Spaces parity and mirrorSlabs allocated across columns, a pool database, and the states Microsoft documents: no redundancy, detached, incomplete→mdadm and LVMSuperblocks at 0.90, 1.0, 1.1 and 1.2, a data offset that moves with the version, event counts, and --create→RAID 1E, 5E and 5EEIBM ServeRAID and Adaptec layouts with the mirror rotated or the spare folded into the stripes→JBOD, spanned and dynamic disksConcatenation, Windows RAID-5 and mirrored volumes, and the LDM database at the end of every dynamic disk→Why the level matters less than what was done to the set.
Every level on this page is recoverable from images of its members, provided enough of them still read and nobody has written the wrong thing over them. What decides a job is rarely the level. It is whether the survivors were rebuilt into a second failure, whether a foreign configuration was cleared to tidy the screen, whether a new array was created over the old one and initialised, whether chkdsk or fsck ran on a volume assembled with the wrong stripe size, and whether the drive that failed first was kept or handed to the courier. Each of those is a page of its own under what the controller says, with the button to leave alone named in the vendor's own words.
The arithmetic is the same on every bench. The controller's metadata, where it survives on the drives, gives the order, the stripe size, the parity rotation and the offset; where it was cleared, parity across the members at the same offset should XOR to zero on a consistent stripe, which proves the level and finds the member that dropped out first; where parity falls says the rotation; the file system's own structures fix the start and the stripe size. Then the set is assembled in software from the images, with nothing written to any of them, and the file system repaired on a copy of the virtual volume. How a RAID array is rebuilt from member images goes through it step by step, and the first-response page says what to do tonight.
The members themselves are the drive site's subject: a clicking 15K SAS drive, a 520-byte drive from a NetApp shelf, a self-encrypting drive separated from its key, an HPE SSD that stopped at 32,768 hours. Those pages are at raiddrivedatarecovery.co.uk, and the same bench images them before the set is touched. A NAS unit as a whole, with its dashboard and its Repair button, is the NAS site's subject at nas-data-recovery.co.uk; the SHR and RAID-Z pages here cover only the array layer inside it.
The questions that come up first.
Which RAID level is easiest to recover?
RAID 1, because there are two whole copies and the job is deciding which is newer. RAID 5 and 6 are routine when enough members read; RAID 0 is the hardest, because there is nothing to check the geometry against but the data itself.
The set is degraded but still working. Should I rebuild?
Only if every surviving member is healthy and you have a backup. If a second drive shows predictive failure, the rebuild reads the bad sectors that finish it. Copy what matters, power down, and image the weak drives first.
Does the level change the price?
No. The number of members does. A set of two to four is £500 + VAT upwards; larger sets, 15K SAS sets, storage shelves and enterprise storage from £1,250 + VAT. Fixed in writing after the free look.
Can I use the calculator to see how risky my rebuild is?
Yes. The RAID calculator gives usable capacity, tolerance and an unrecoverable-read-error estimate for a rebuild, with the caveats the estimate needs.
Not sure which level it was?
Photograph the controller's screen or export its log and send it with the form. The metadata on the drives says the rest, and the first look is free whichever level it turns out to be.