Far more of these sets are finished off during the rescue than by the fault that started it: another reconstruction loaded onto worn disks, or a menu item that flattens the very records a rebuild has to read. This work takes that gamble away. Every member is copied before anything else is attempted, the geometry is proved rather than assumed, and nothing is ever written back to the disks you sent us.
A RAID 5 is quoted, never guessed at. Diagnosis costs nothing and the written figure follows it — from £500 + VAT counted by member, logical rebuilds keeping no fix, no fee.
Losing one disk is the exact event this layout was designed to absorb, and on its own it rarely ends in a recovery job. What fills the diary is the six hours afterwards, and Merseyside server rooms manage to reach all six of the positions below.
The replacement goes in and the controller settles down to read every surviving member from end to end, which on modern capacities means hours of continuous work asked of disks that are the same age as the one that just died. One of them gives out. A set that was merely degraded before lunch is gone by the time the office closes, and nothing reaches this bench more often.
Backplanes, host adapters and controller cards fail on their own schedule, and it is frequently years ahead of the disks they were bought to look after. Not a single member is faulty and the volume still never appears again, because the description of how it was assembled died with the hardware.
Any disk that fell out a fortnight ago is holding a fortnight-old photograph of the set. Bring it back online and the controller begins folding that stale picture into live stripes, corrupting parity across the whole set as it goes, and doing it quietly.
One duff sector on one member is exactly what parity exists to paper over. Two duff sectors at the same stripe address on two members is a hole arithmetic cannot fill, which is why every weak region on every disk gets imaged around before any stripe is calculated.
Menus offer re-create, initialise and repair with warnings that do not begin to convey what they do, and any of the three can overwrite the metadata describing where your data actually lives. Where one has already been run, that is precisely the point to stop.
Adding a disk, or converting from one layout to another, is many hours of continuous rewriting. Interrupt it with a power cut, a crash or somebody running out of patience, and the set is left stranded between two geometries, valid as neither.
No disk of yours is ever put back into a running array here. Work opens with each member going onto a hardware imager, and any disk carrying a physical fault visits the clean bench before that happens. Everything from then on is done against the copies. Which disk sat where, how large the blocks were, which way the parity rotates and where the data starts are all deduced from those images; the volume is then assembled virtually on top of them, and your files are lifted off that assembly. The original disks sit on a shelf, untouched, as a fallback nobody expects to need.
Dropping in a spare and giving the box another go has the feel of decisive action about it. On a set whose single disk of redundancy is already spent, it is close to the most dangerous option on the menu. Reconstruction puts its heaviest demands on the tiredest members at the exact moment none of them can absorb another problem, and a second casualty converts a straightforward job into a slow, delicate one. Read errors piling up in the log, or a reconstruction that has ground to a halt, both mean the same thing: take the power off. Standing still costs an array nothing at all.
Much of what arrives here is the entire trading record of a small firm: Dell PowerEdge and HP ProLiant boxes, towers somebody assembled themselves, Synology, QNAP and Netgear units, coming from offices in the city centre and the Baltic Triangle, from the estates at Speke and Knowsley, from the yards at Bootle and from across the water on the Wirral. The disks go to Manchester Data Recovery, Peter House, Oxford Street, Manchester M1 5AN, about 35 miles along the M62 or a working day by tracked post, and a set holding up a business can be moved forward in the bench queue. Everything outside this layout is covered by RAID recovery and NAS recovery, and the figures behind all of it are on the data recovery cost page.
Parity is not a spare copy. For every stripe written across the set, the controller works out one block that is the running total of the others, and stores it on whichever member is next in the rotation. Lose any single block and the missing one can be calculated back from everything that survives. That is the whole trick, and it is an elegant one, but notice what it buys: exactly one missing piece per stripe. A second gap in the same stripe leaves the sum unsolvable, and no amount of controller firmware changes that.
The trouble is that disks bought together tend to fail together. They came off the same production run, went into service on the same afternoon, and have spent every hour since doing identical work at an identical temperature. When one wears out, the others are not bystanders; they are the same age with the same mileage. That is the position a rebuild puts them in, and it is why the second failure so often arrives during the first recovery.
To reconstruct a replaced member, the controller has to read every sector of every surviving disk. Not a sample: all of it. On a set of four four-terabyte disks that is twelve terabytes of continuous sequential reading, which takes many hours and represents by far the heaviest sustained load those disks have ever been asked to carry. Any weak sector that has been sitting there unnoticed, and a disk that has been running for years will have some, gets found during that pass. Found during a rebuild, on a set with no redundancy left, it is not a warning. It is the end of the rebuild.
Which is why the advice on this page is so consistently to stop. A degraded array that is switched off is a set of disks in exactly the condition they were in when the first one died, and every one of them can be imaged gently, one at a time, with as many retries as a weak region needs. A degraded array that is rebuilding is a set of disks being pushed hard with nothing to catch them.
Once the images exist, the set has to be described before it can be read, and four things have to be established. The order the members sat in, which the bay labels often disagree with. The block size, which varies by controller and by whoever configured it. The rotation, meaning where the parity block moves to on each successive stripe, and whether it moves left or right, in step with the data or offset from it. And the start offset, since most controllers reserve space at the front of each disk for their own metadata and the data begins somewhere after it.
Get any one of those wrong and the reassembled volume reads as convincing noise: a partition table that looks plausible, directories that open, files that are corrupt. So the parameters are not guessed at, they are proved. Known file structures are located across the images and their fragments matched up until only one description of the set can possibly be correct. Then the file system is raised on top, and only then does anybody start talking about which folders are coming home.
Not every RAID 5 is a card in a server. Intel RST on a desktop motherboard, mdadm on Linux, Storage Spaces on Windows and the vendor layouts inside Synology and QNAP units all produce sets that behave like RAID 5 and describe themselves quite differently. Some keep their metadata at the front of each disk, some at the back, and some spread across both. Several use a modified geometry that is not textbook RAID 5 at all, which matters enormously to the reassembly and not in the least to how the disks should be handled before they get here.
Whatever the flavour, the instructions are identical. Power off, label the bay each disk came out of if you possibly can, keep the set together in one box, and note the make and model of the controller or the unit. Those two minutes of writing things down save hours at the bench and occasionally decide whether a marginal set comes back whole.
Cut the power, box every member including the dead one, and the set gets solved in software. Figure first.
Nearly everything on this bench arrived by tracked, insured post. It is the calmest way to move a failing drive, and a parcel handed in anywhere on Merseyside is normally on the bench the next working day.
Still screwed inside a laptop, tower, MacBook, iMac, server or a CCTV recorder? Take the drive out and post that on its own. We do not strip machines here, and any computer shop will do it in a few minutes. The one job nobody can take on is flash soldered straight to a logic board, as on Apple Silicon Macs and a handful of very slim laptops: if the storage will not unbolt, there is nothing to send.
↓ Print the shipping & booking-in form (PDF)
Address it to Manchester Data Recovery. It is roughly 35 miles from Liverpool along the M62 if you fancy the run, and next working day by tracked post if you do not. We ring or email the moment it is booked in, and the free diagnostic closes 2 working days after that.
Not sure what belongs in the parcel? Call 0800 689 0668 before you tape it shut, or step through the free online diagnostic first.