Open, and the first look is free · 9am–5:30pm, Mon to Fri Quicker to ring: 0800 6890668
RDR Reading Data Recovery 0800 6890668 Get it quoted
RDR / What we take in / RAID 5 arrays and rebuilds

RAID 5 · one parity drive

RAID 5 data recovery, Reading. Copy every member before anyone rebuilds. Usually that alone decides it.

Parity in RAID 5 protects you against losing a single disk and nothing beyond it. If you lose a second disk, most likely it will be during the rebuild, because a rebuild needs a complete, uninterrupted read from drives bought in one carton and run at the same temperature ever since. Disks reach the Oxford intake lab by post from distribution centres at Theale, small software teams at Earley, and the four-bay box in the stores at a Woodley trade counter. The front panel nearly always reads worse than what we find, because failed is a decision made by a controller, not a diagnosis. From £500 + VAT, one figure in writing after a free look, and the disks you send go home exactly as they came.

Only pay if we recover your data, on most jobs The diagnosis is free; then one fixed figure in writing Parcels arrive from Newbury, Woodley and Theale

An engineer answers, and looking is free
0800 6890668

Match the RAID 5 symptom to the fault.

Different fault? Begin at the triage →
What you are seeingWhat is likely wrongWhat happens next
The controller reports a member missingNow your one disk of parity is used up, and a second failure takes out the volumeShut it down before another member fails
The rebuild ran most of the night, then quitA full read of the other disks hit a soft patch on another disk of the same age: the usual endingKeep it switched off. Do not retry
Two members are showing as failedGenerally a finished disk next to one that is slow on just a few sectors: they are not the same kind of faultPost them all; the images separate them
A rejected disk was forced back into the arrayOut-of-date blocks now sit among live stripesTouch nothing else
A replacement controller cannot see the setThis is an argument over metadata; not one of your files has changed locationA normal reassembly
Recovery software has been run on the individual membersData has been written to your original disksSend the disks regardless, and tell us which tool was used
Packing and posting: first, get a quote either by phone or with the quote form. Next, make sure all disks are labelled with their bays, then wrap them so they cannot move within the box. Put a value on the parcel that represents the data on the disks (not the hardware) and send it tracked and insured to the Oxford intake lab, or drop it off at the Oxford intake lab, Monday to Friday, 9am to 5:30pm, no appointment needed. We pay for the return postage. If you wish to talk the packing over with an engineer before sealing the box, call us first. All of this is also on our packing and postage page.

What a RAID 5 rebuild demands of the disks that are left.

What a rebuild has to get throughA rebuild runs at one rate, which is flat out. Each remaining member is read from beginning to end, so with four 16TB disks in a chassis that means about 48TB out of the three that are left, and each of them carries the same warm running hours as the member that stopped.
What a URE rating tells youConsumer drives are specified for one unreadable sector per 1014 bits, roughly one awkward sector in every 12.5TB. The enterprise specification is 1015, ten times the headroom. Consider this the worst a manufacturer will put its name to, not a countdown. Many disks still in service easily run past these numbers. The figure is sound. The frightening graphs drawn on top of it are not.
What the controller doesAge decides it. An older controller loses all progress up to that point when it meets a sector it can't complete. Newer PERC and MegaRAID firmware will punch out the stripe in question and continue; mdadm under Linux records the bad block and does the same thing. It is always better to lose one stripe to one unreadable sector than to lose the entire volume.
Why the next disk stopsOne carton, one chassis, one temperature history between them. Ask every disk for a massive read all at once, and whichever is quietly failing will stop there. Instead we go over the same ground slowly and leave the soft patches until last.

From your parcel to your files coming back.

01

Logged the day it lands, and the look costs nothing Free

A case number is assigned to your device when we receive it. An engineer then finds the real fault (free) before anything else occurs; you will be advised clearly of what will be returned and what will not, along with a single written price. Charges cannot begin until you have read this and accepted.

The look costs you nothingA fixed price, in writingStill nothing to pay
02

Image each member first

Every member of your set is copied completely to our storage, including those your card wrote off. Slow reads are logged as they happen and the worst regions are left till last. Not one byte goes back to the disks you posted.

Every member cloned firstThe rejected members too
03

Take the geometry off the members

There are three things that need to be determined: which bay each member came out of, how big a stripe is, and which way parity travels round the set. All three come from what each disk recorded about itself, checked against the pattern of your own data. Nothing here is estimated.

Bay order and stripe size settledChecked before assembly starts
04

Assemble the array over the images

Then the array is put together on top of the images, never on the disks. Each block comes from whichever copy delivered it most cleanly; where no copy did, parity fills in the missing data. That is why data on a member the controller had written off still arrives home. Once the file system has been repaired, the rebuilt volume is mounted and checked before anything is released.

Every block from its clearest copyMounted and tested
05

The list first, then the bill, then the files

You look at the file list first and make a decision about what to do. We will never invoice anything until you have said yes, and in most cases you only pay if we recover your data. All of the information that we recover will be put on new media purchased by us specifically for this job, and our company pays the cost of postage back to you. The case stays open on the bench until you tell us the files open on your own machine.

Nothing billed until you say yesNew media, bought for the jobPostage home is on us

What goes wrong most often

  • Degraded describes something that has already happened: the set was carrying one disk of cover, and when the member left it took that cover with it. From then on the array is as exposed as a single drive.
  • Fear sells array recovery: this is the most expensive work any bench does, and a lot of what is written about it is intended to unsettle the reader. Instead of selling you fear, we tell you what the imager is finding on your own disks.
  • A puncture costs a stripe, and only a stripe: modern firmware that encounters an unreadable sector marks that stripe, logs it and continues with the rebuild, whereas older cards abandoned the entire rebuild on the spot. With the set assembled, punctured stripes can be listed for you by address.
  • Linux software arrays keep records: mdadm's superblocks, read next to the bad-block log, show which disk stepped out of the array and roughly when. Most hardware RAID controllers say a great deal less.

The arithmetic, done once: a consumer disk is rated at one unrecoverable read error (one bad sector) in every 1014 bits, which means about one awkward sector every 12.5TB. Enterprise disks are ten times better, at 1015. With four 16TB drives in a chassis, a rebuild needs roughly 48TB of faultless reading from the remaining three, so this is arithmetic rather than scaremongering. Set against it, disks in the field routinely run well past what they are rated for. Both point the same way: image first, then rebuild, since there's no cost to a rebuild once you've made your images.

How one of these actually went.

RG · RDG-2026-1972ON FILE ✓

Two RAID 5 members gone, and the year-end still filed on time

It sounded like two of the disks had failed at once. They had not. One member of the array had quietly left the set months before without anyone noticing, and a second stopped midway through the rebuild that followed. Imaging found only nineteen bad sectors on that second disk, so three good members and one almost as good were enough between them to work the stripe out again. The year-end filing went in on time.

19 unreadable sectors, not a second disk failure6 days from arrival to return

What helps, and what makes it worse.

First things to do

  • Switch off the chassis the moment a member leaves
  • Write the bay position on each disk before it comes out of the chassis
  • Include each member, even the written-off ones
  • Say which controller it is, and what has been done so far

What makes it worse

  • Starting or resuming a rebuild while a member is missing
  • Using the card's menu to force a rejected disk back online
  • Pointing consumer recovery software at the members
  • Fitting a spare and letting the controller initialise

Answers before you commit to anything.

The card shows two members failed. Is the volume gone?

Usually not. Controllers drop members for reasons of their own, and a few sectors taking a little longer than usual to answer will do it. Dropping a disk from an array does not mean the disk is finished. We image every member of the set before we do anything else, then use the parity the other disks still hold to reconstruct whatever a single image is missing. So a controller marking two disks bad is just part of an ordinary week's work; two disks genuinely beyond imaging is rare.

We are running degraded with a member missing. Do we rebuild?

Not until the copies exist. A rebuild forces every remaining member to read itself end to end, an operation an exhausted disk cannot meet, and it lays new parity across the existing parity that could have reconstructed your volume. Once you've imaged the set you can rebuild as many times as you want, because nothing you do afterwards will affect the copies.

Do the URE numbers actually matter?

It is a sector that the drive's internal error correction cannot put right, at which point the drive stops and reports that it has done so, rather than passing back incorrect data. Consumer drives are specified at roughly one such error every 12.5TB moved; enterprise drives one every 125TB. Take these numbers as minimums a manufacturer will defend, not as a date in your diary for when to expect a URE. Modern array firmware handles a URE by punching out that single stripe and carrying on.

Should the controller come with the disks?

No: the disks themselves are sufficient. Each member contains its own account of how the set was built, so when that account is damaged, the shape of your own data says the same thing. The set is then assembled in software above the images, nothing is written to what you posted, and your controller stays in the machine.

If this sounds like yours, start here.

Servers and RAID arrays → Servers that will not boot → Nothing shows up in Windows → RAID 1 mirrors → Dell PowerEdge servers → Every device we handle →

Nothing gets worse while it sits switched off.

There is no charge for the look. You then receive a list of the files that opened, those that did not, and a fixed amount to complete the job. Keep it switched off until you have received that in writing.

0800 6890668

Get it looked at Posting it to Oxford