Open, and looking at it costs nothing · 9am–5:30pm, Mon to Fri Ring instead, it is quicker: 0115 8220606
NDR Nottingham Data Recovery 0115 8220606 Get my quote
NDR / Devices and media / RAID 5 arrays and failed rebuilds

RAID 5 · one disk of parity, and no more

RAID 5 recovery, Nottingham. Clone every member before anybody rebuilds. More often than not, that decides it.

Parity on a RAID 5 buys cover for exactly one absent disk. The thing that finishes the second, far more often than a fault does, is the rebuild — a single flat-out read from end to end, demanded of drives that shipped in one carton and have shared one cabinet temperature ever since. Price first, because nobody else publishes it: from £500 + VAT for arrays, NAS boxes and servers, with the figure fixed in writing once the free look is done, and the disks you post go home untouched. Ignore the front panel, too. Failed is a verdict a card reaches in a second, not a diagnosis. Sets come in from Nottingham and across the East Midlands: a QA office at a Worksop food plant, a small software outfit in Chesterfield, and the four-bay box behind a Lincoln trade counter.

No files back, no bill — on most jobs Free diagnosis first, then a single price, in writing Parcels come in from Derby, Lincoln and Loughborough

Ask an engineer, and the first look is free
0115 8220606

What a RAID 5 symptom points to.

Not yours? Open the fault finder →
What is happeningWhat usually causes itWhat that means for you
The card says a member has goneYour one disk of parity is spent, and a second fault now takes the volume with itStop it now, before a second disk goes
A rebuild went half the night before it haltedOne disk of the same age hit a soft patch during that full read — the ending we see mostLeave it powered down. Do not try again
Two members reported failedNearly always one finished disk beside one that is merely slow over a few sectors, rather than a matched pairPost the lot; the clones will tell them apart
A thrown-out disk was forced back inStale blocks now sit in among live stripesTouch nothing further
The new controller finds no arrayA quarrel about metadata, that is. Your files have not moved an inchA routine reassembly job
Recovery software has been pointed at the membersSomething has been written to the disks you still havePost them anyway, and tell us which tool ran
How to pack and post it: box it so that nothing can shift about, and insure the parcel for what the files are worth rather than the drive. Tracked post to the intake lab, please; we pay for the journey back. Not sure what to put in the box? Ring before you tape it up and an engineer will tell you. The whole of it is written out on the packing and postage guide.

What a RAID 5 rebuild demands of the disks still in the chassis.

How much a rebuild readsOne speed is all a rebuild has, and that speed is flat out. Every disk still in the chassis is read from end to end: put four 16TB members in and the three survivors owe you about 48TB between them, each of them with a running-hours count matching the disk that stopped.
Reading the URE specificationConsumer drives carry a specification of one unreadable sector in 1014 bits, or roughly one difficult sector in every 12.5TB. The enterprise rating is 1015, ten times the headroom. Read that as the worst a manufacturer will put its name to and not as a countdown, then notice how many disks in service run far past it. Nothing is wrong with the figure. The frightening graphs built on top of it are another matter.
How the card answersAge decides it. An older controller that meets a sector it cannot complete throws away every hour the rebuild has already put in. Firmware of a newer vintage, on PERC and MegaRAID, marks the stripe as punctured and keeps going; mdadm on Linux notes the bad block and does likewise. Surrendering one stripe over a single unreadable sector is a price worth paying. A whole volume given up to it never was.
Why the next disk followsOne carton, one chassis, one shared temperature history. Demand a single enormous uninterrupted read of every disk and whichever of them was quietly failing gives out there and then. Here the same ground is covered slowly, with the soft patches held back to the end.

From the parcel arriving to the files going home.

Jobs already finished →
01

Booked in the day it arrives, and the first look is on us Free

Your device picks up a case number the day it arrives, and an engineer then establishes what has genuinely gone wrong — at no cost, and ahead of everything else. Two things come back to you together: a plain account of what can be lifted and what cannot, and one fixed figure in writing. Agree to that figure, or turn it down and pay nothing.

The first look is freeOne fixed figure, in writingNo charge at this point
02

Clone every member first

A full copy of every member goes onto our own storage, and that includes the ones the card wrote off. Where a read comes back slowly it is logged at the time, and the worst regions wait their turn until the end. Not a byte travels back to the disks you posted.

Each member imaged firstIncluding the ones the card threw out
03

The layout taken off the disks

Three questions have to be settled before anything is assembled: which bay each member came out of, how large a stripe is, and which way round parity rotates. Each answer is taken from what the disks recorded about themselves, then checked against the shape of your own data. Nothing at that stage is estimated.

Bay order and stripe size worked outConfirmed before anything is assembled
04

Assemble the array above the clones

Assembly happens on top of the images and never on your disks. Whichever clone gave a block up most cleanly is where that block comes from; where none of them did, parity fills the gap — which is how data living on a member the card had written off still reaches you. The file system is repaired after that, and the volume mounted and checked before a thing is released.

Each block from its cleanest cloneVolume mounted and checked
05

The file list comes before the bill

The file list reaches you before any invoice does. Say yes and it is billed; say no and it is not — and on most jobs, if nothing comes back there is nothing to pay. Whatever comes off goes home on media bought in for your case, carriage at our end. Nothing here is closed until you have opened those files on your own machine.

Nothing is charged until you agree the figureNew media, bought in for your jobWe pay to send it home

What comes in most often

  • Degraded describes something that has already happened — one disk of cover was carried, and the member that left took it away. From that moment on, the array stands as exposed as a single drive with nothing behind it.
  • Fear sells array recovery — no bench does dearer work than this one, and a fair proportion of what gets written about it is there to unsettle you. Here you get instead whatever the imager actually finds on your own disks, put in writing.
  • A puncture costs one stripe, and one stripe only — meeting a sector it cannot read, current firmware notes the stripe, logs it, then presses on; an older card would throw the entire rebuild away on the spot. With the array assembled, every punctured stripe can be listed out for you by address.
  • Linux software arrays keep records — mdadm superblocks, set beside the bad-block log, will show which disk stepped out of the array and roughly when it happened. Hardware cards tell you a great deal less.

The arithmetic, done once: what a consumer disk is rated for is one unrecoverable read error across 1014 bits, coming out at a sector every 12.5TB or thereabouts, while enterprise disks are rated ten times better than that, 1015. Fit four 16TB drives and a rebuild will want something in the region of 48TB read faultlessly off the three still standing, so this is arithmetic rather than scaremongering. Against it, disks in service routinely run far past what they are rated for. Both statements hold, and both argue for one order of work: clones, then the rebuild — because a rebuild after cloning costs you nothing.

One of these, from start to finish.

NG · NTG-2026-1972ON THE LOG ✓

The year-end still went in on time, with two RAID 5 members gone

Nineteen bad sectors, and that was the whole of it. The set looked like it had gone two disks down together; in fact one member had dropped out quietly months earlier with nobody the wiser, so when a second stopped partway through the rebuild there was nothing left holding it up. Imaging the pair showed how little on that second disk was actually unreadable, and three sound members plus one very nearly sound worked the stripe out between them. Year-end filing still went in on the day.

19 unreadable sectors, and not a second dead disk6 days between arrival and return

What helps, and what does damage.

Worth doing first

  • Cut the chassis power the moment a member drops out
  • Write each disk's bay position on it before removal
  • Include the whole set, written-off members as well
  • Name the controller, and say what has been attempted

What makes it harder

  • Starting a rebuild while a disk is missing
  • Forcing a rejected disk back online from the card's menu
  • Aiming off-the-shelf software at the disks
  • Fitting a spare and letting the card initialise

Answers before you spend anything.

Two members are showing as failed on the card. Is the volume gone?

Seldom. A controller will throw a member out for reasons entirely its own — a handful of sectors answering a fraction late is quite enough — and being dropped from an array is a very different thing from being finished. Before anything else is attempted, every member is cloned onto our own storage; parity held by the rest then supplies whatever a single clone will not give up. Two members flagged bad by a card makes for an ordinary week here. Two genuinely beyond cloning is rare.

We are running degraded, a member short. Do we rebuild?

Not until clones exist. Every disk still in the chassis is told by a rebuild to read itself end to end, precisely what a tired drive cannot manage, and fresh parity is laid down on top of the old — when it was that old parity which would have reassembled your volume for you. Clone the set first and you may rebuild as often as you like afterwards, since nothing done later can reach a clone.

Do the URE figures actually matter?

A URE is a single sector that a drive's own error correction is unable to fix; the drive then stops and reports it rather than hand back something wrong. Consumer specifications work out at about one such event for every 12.5TB read, and the enterprise figure is one in 125TB. Read them as the least a manufacturer is prepared to stand behind, not as some countdown in the diary. Where firmware is modern, meeting a URE marks the stripe as punctured and the rebuild continues.

Does the controller have to come with the disks?

No — the disks by themselves will do. How the set was built is recorded on every member, and where that record has been damaged, the shape of your own data tells us the same thing. Assembly then happens in software above the clones, your originals are never written to, and the card stays bolted where it is.

Start here if the same thing is happening to yours.

Server and RAID work → When the server will not start → The disk is not detected → The full device list →

Switched off, a drive cannot get any worse.

The look costs nothing. What comes back to you is a file list — what opened, what did not — plus one figure to finish the job, in writing, before anything chargeable begins. Most jobs carry no fee at all unless the data comes back. Leave the drive switched off until you have that list.

0115 8220606

Send it for a free look Posting it to Manchester