What finishes off a parity set is usually the attempt to save it: a second full rebuild demanded of tired discs, or a menu option that wipes the very description a reconstruction needs. None of that happens here. Each member is copied before anything is tried, the layout is worked out from those copies and demonstrated rather than assumed, and not one byte goes back onto the discs you posted.
A parity set is quoted, never guessed at. The diagnostic costs nothing and the written figure follows it — from £500 + VAT counted by disc, with logical work under no fix, no fee.
One dead member is the exact event this layout was bought to absorb, and by itself it seldom ends in a recovery job. What fills the diary is the afternoon that follows, and these six between them account for nearly every parity set that reaches this bench.
The spare goes in and the controller settles down to read every remaining disc from one end to the other. On today's capacities that is hours of unbroken work demanded of drives the same age as the one that just died, and one of them declines. A set that was merely degraded at lunchtime is beyond the controller by close of business.
Backplanes, host adapters and controller cards keep their own schedule and it often runs years ahead of the drives they were bought to look after. Not one member is faulty, the volume never appears again, and the reason is that the description of how it was assembled died with the hardware.
A member that dropped out three weeks ago is holding a three-week-old photograph of the set. Bring it back online and the controller starts folding that old picture into live stripes, poisoning parity across the whole array as it goes, and doing it without a word.
A single unreadable sector on one disc is precisely what parity exists to cover. Two unreadable sectors at the same stripe address on two discs is a gap the arithmetic cannot close, which is why every soft region on every member is imaged around before a stripe is calculated.
Controller menus offer initialise, re-create and repair behind warnings that do not begin to describe what they do, and any of the three can flatten the metadata recording where the data actually lives. If one has already been run, that is the moment to stop rather than the moment to try the next one.
Adding a disc, or migrating from one level to another, means hours of continuous rewriting. Interrupt it with a power cut, a crash or an engineer who ran out of patience, and the set is stranded between two geometries and valid as neither.
Parity is often explained as a spare copy of the data, which it is not. For each stripe written across the set the controller works out one extra block from the others — a running total, in effect — and puts it on whichever member's turn it is. If one block in that stripe later goes missing, the total plus the surviving blocks are enough to calculate what was lost. Elegant, cheap in space, and strictly limited: it will cover one absent block per stripe and no more. Two gaps in the same stripe and the sum has more unknowns than equations, which is a fact about arithmetic rather than about firmware.
What makes the arrangement fragile in practice has nothing to do with the mathematics. Discs bought together fail together. They came off one production run, went into service on the same afternoon, and have since done identical work at an identical temperature in the same warm cupboard. When one wears out the others are not innocent bystanders — they are the same age with the same mileage, and they are about to be asked for the hardest hours of their lives.
Rebuilding a replaced member means reading everything the other discs hold — not a sample, not the used portion, every sector. Four four-terabyte drives make that twelve terabytes of continuous sequential work, comfortably the hardest thing those drives have ever been asked to do. Weak sectors that have been sitting unnoticed in a rarely visited corner, and any drive with a few years of power-on hours has some, are found during that sweep. Discovered mid-rebuild on a set with no redundancy left, such a sector produces no warning to act on. It ends the rebuild, and frequently the volume with it.
There is a quieter version of the same problem, and it comes off the specification sheet. Consumer drives are sold with an unrecoverable read error rate of about one sector in every hundred trillion bits. In daily use that number is invisible. Ask for twelve terabytes in a single uninterrupted sweep and it stops being a footnote and becomes a reasonable expectation. Hence the advice repeated on this page: an array that has been powered down is simply a group of discs in the same state they reached when the first one died, and each of them can then be copied slowly, on its own, with as many retries as a soft patch happens to need.
Most of what arrives here is the whole trading record of a firm that never expected to be a recovery customer: Dell PowerEdge and HP ProLiant boxes, towers somebody built themselves, Synology, QNAP and Netgear units. They come from the business parks along the A329(M) and Western Road, from the software and instrument firms on the M4 corridor between Reading and Slough, from engineering units around Camberley and Frimley, and by tracked post and courier from every London postcode on the London page. The discs travel to Guildford Data Recovery, Building 2, Ground Floor, Guildford Business Park, Guildford GU2 8XH — forty minutes from Bracknell on the A322 and the A3, or one working day in the post. Other layouts are covered on RAID recovery and NAS recovery, and every figure behind them is on the data recovery cost page.
No disc of yours is ever returned to a running array in this lab. Work opens with each member going onto a hardware imager, and any drive carrying a physical fault visits the clean bench before that happens. Everything from that point is done against the copies. The originals go on a shelf untouched, as a fallback nobody expects to need, and the reconstruction is free to be attempted, discarded and attempted again without costing anything except time.
A pile of images is not yet a volume. Four properties have to be pinned down before anything can be read out of them. Which slot each member occupied, since the labels on the bays and the truth inside the metadata part company more often than anyone expects. How big the blocks are, which depends on the controller and on whoever set the box up years ago. How the parity travels — left or right through the rotation, synchronous with the data or displaced from it. And how far into each disc the data begins, because most controllers keep the first part of every member for their own bookkeeping.
Get one of those four wrong and the failure is quiet. Up comes a partition table, a directory tree, files with sensible names and dates, and contents that are gibberish. Nothing warns you. That is why none of the four is entered on a hunch here: each is demonstrated. Structures with a known shape are found across the images, their fragments matched against one another, and every candidate description tested until a single one survives. The file system is raised on that description, and only after real files open correctly does anyone start talking about which folders are coming home.
Not every RAID 5 lives on a card in a rack. Intel RST on a desktop board, mdadm on Linux, Windows Storage Spaces and the vendor layouts inside Synology and QNAP units all behave like RAID 5 and describe themselves in entirely different ways. Some keep metadata at the front of each disc, some at the back, some in both places. Several use a modified geometry that is not textbook parity at all. All of it is readable; none of it is readable by guessing, and it makes no difference whatever to how the discs should be handled before they get here.
Power the box down and leave it down. Do not authorise a rebuild, and do not accept any menu item offering to initialise, re-create or repair the array. Label each disc with the bay it came out of, because bay order is the single most useful thing you can hand over and the easiest to lose. Keep the set together in one carton, including the member that failed first — it is frequently the most useful disc in the box. Note the make and model of the controller or the NAS, and if a dialogue appeared and somebody clicked through it, say so on the booking-in form. Two minutes of writing things down saves hours at the bench, and now and then it decides whether a marginal set comes back whole.
Power off, box every disc including the dead one, and let the set be rebuilt in software instead.
Hardly anything walks into this lab; it arrives in a padded box. Tracked, insured post is the least stressful thing that can happen to storage that has already given up, and a parcel handed over at a Bracknell post office before the last collection is usually in Guildford the following working day.
Is the storage still bolted into a machine — laptop, tower, iMac, MacBook, rack server, a DVR under the till? Free it first and send the bare unit. Stripping hardware is not something this lab does, though it is ten minutes' work for any repair shop on your high street. There is a single case with no way round it: memory chips soldered flat onto the mainboard, which is how Apple Silicon machines and certain ultrabooks are built. Where the storage cannot be unbolted, there is no parcel to make up.
↓ Print the shipping & booking-in form (PDF)
The name on the parcel wants to be Guildford Data Recovery. Driving it over from Bracknell is roughly forty minutes on the A322 then the A3; posting it costs you a stamp and a day. Either way, a message goes out to you as soon as it is logged onto the system, and two working days later the diagnostic is finished.
Unsure whether something should go in the box? Ring 0800 689 0668 while the lid is still open, or work through the free online diagnostic and let it tell you.