ADR / Media and machines / RAID 5 sets and rebuilds that stopped

RAID 5 · a single disk of parity

RAID 5 data recovery, London. Image every member before anyone rebuilds. That decision decides whether the volume comes home.

The parity for a RAID 5 protects against the loss of one complete physical drive. What ends the second is usually not a fault at all; it is the rebuild: a single flat-out read from beginning to end, on drives that are from the same original carton and have been in the same environment ever since. The cost comes first because few people publish this number: £500 + VAT is where the price for an array, NAS or server starts, and the exact amount is written down after an engineer has had a look, free of charge. The disks you post to us will go home unaltered. Do not even bother looking at the front panel: in about a second the RAID card will decide it has failed; however, this does not constitute a diagnosis. Sets come from a Harley Street practice using a basement NAS for its image storage, from a Park Royal food processing company storing batch and traceability data, and from a Barnet accounting firm storing its files on a four-bay box located in its rear office.

Only pay if we recover your data, on most jobs A free look first, then one figure in writing Media posted in from Croydon, Ealing and Camden

Put it to an engineer — the first look costs nothing
0203 0868633

RAID 5: what the symptom usually means.

Different fault? Try the finder →

What is going on	The usual reason for it	Where that leaves you
The card detects that one of its members has disappeared	The single disk of parity is spent, and the volume is lost with the next fault	Remove power before another member disk fails
A rebuild ran half the night and then stopped	A disk of the same age hit a soft patch during that full read, which accounts for most rebuild failures.	Don't switch it back on, and don't try again.
Two members marked failed	Almost invariably you'll have one finished disk adjacent to one that's just slow over a small number of sectors, not a matching pair.	Ask first for a quote, then send us the whole set; the images will separate them.
One of your dropped disks got pushed back in	You've got stale blocks mixed in with live stripes now	Do nothing else to it
When you install a new controller, it sees no array	That is a disagreement over metadata. None of your files moved anywhere.	Reassembly, and nothing worse
Recovery software has been directed at your members	Writes have gone onto the disks you're keeping	Ask for a quote first, then send them anyway, and tell us what tool ran.

Packing it and posting it: get a quote first (by phone or with the form), then box it so nothing can move about. Send it tracked and insured to our Guildford intake lab, or drop it off there, Monday to Friday, 9am to 5:30pm (no appointment needed); the leg home is paid at this end. Want a word first about what goes into the box? Ring us before it is sealed and an engineer will run through it. You will find the lot spelled out on the guide to packing and posting.

What a rebuild asks of the disks left in the chassis.

How much a rebuild has to readThere's only one speed at which a rebuild takes place: flat out. Therefore, every disk left in your chassis gets read straight through: four 16TB members means that the three still good ones owe approximately 48TB between them, each carrying a running-hours count that corresponds to the disk that died.

How to read the URE figureConsumer specifications call for one unreadable sector per 10¹⁴ bits (or roughly one troublesome sector per 12.5TB). Enterprise kits specify 10¹⁵, which is ten times the headroom. Read it as the minimum performance level that a vendor is willing to stand behind rather than as some sort of countdown, and note how many drives in service run far beyond that. There's nothing wrong with the number itself. What has been built on top of it in graphical format is a separate matter.

What the card does about itIt depends on how old your card is. Older controllers will discard every hour of rebuild effort already completed whenever they encounter a sector they cannot finish reading. Newer versions of firmware on PERC and MegaRAID mark those stripes bad and continue onward; mdadm under Linux will record the bad block and carry on in the same way. Losing one stripe for every unreadable sector is worthwhile; losing an entire volume never was.

Why the next one followsOne carton, one chassis, one shared history of temperature. Ask for one long continuous read of all disks, and whichever one was quietly failing will fail immediately. The same basic territory is covered slowly here, with the soft patches intentionally left until last.

From the box arriving to your files going back.

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

As soon as your parcel arrives here, an engineer assigns a case number that same day, prior to determining what has really failed, which is the first thing done and costs you nothing. In addition to a note of what can and cannot be recovered from your drive, you will receive at the same time the fixed, written price for restoring what can be recovered. You can accept this quote, or decline it, which will result in no further obligation on your part and nothing owed to us.

Nothing to pay for lookingA single figure, put in writingNothing owed at this stage

Take an image of every member

Each member gets copied whole to our storage (including any that were rejected by your card). Any read which returns slowly gets logged while it happens; the roughest areas get saved until last. No writes whatever go onto your posted disks.

An image of each member firstMembers the controller had dropped included

The layout read off the disks

Three things are determined before assembly: which bay did each member occupy? How large is each stripe? And what is the parity rotation direction? All answers come from information recorded on the drives themselves and are tested against where your data actually lives. Guesswork plays no role here.

Bay order and stripe size establishedSettled before anything goes together

Build the array over the images

Your set goes together on top of those images, never on your original hard disks. Each block comes from whichever image provided it most cleanly; parity supplies anything missing (which is why material still sitting on a rejected member disk will eventually make its way back to you). Once repairs to your file system have completed, your volume will be mounted and read through before we release anything.

Every block from its cleanest imageThe volume mounted and checked

You see the file list before you pay

You will see what has been recovered as a list first, after which a bill exists. Accepting the list means that a bill will be issued. Rejecting the list means that a bill will not be issued, and in most cases you only pay if we recover your data. The recovered data travels from here back to you via new media purchased specifically for your project, with the cost of postage paid by us. Your case is not resolved (i.e., "closed") until you personally view the files on a computer of your own.

No charge until you accept the figureFresh media, supplied with the jobThe post home is ours

What arrives most often

Degraded refers to a state of affairs that describes events in the past: there existed a single disk of cover, and the member that left took that cover off with it when it failed. Since that time the array has been exactly as exposed as a single drive without anything standing behind it.
Array recovery is sold based upon fear: this is the most expensive type of service performed by any engineer on their bench, and much of what gets written about array recovery is designed to unsettle you. What you get here is whatever is found on your own drives by the imager, written down.
A puncture loses only one stripe: on reaching a sector that cannot be read, today's firmware marks that stripe punctured, logs it and continues; older cards would discard the entire rebuild effort at the moment the failure was encountered. Once your array is assembled, all stripes that were punctured can be listed out for you by address.
Linux software arrays keep track: reading the mdadm superblocks alongside the bad-block log will show you which disk left the array and approximately when. Hardware cards typically provide far less info than that.

The arithmetic, once: the consumer rating is one read error that the drive cannot correct in 10¹⁴ bits, or roughly one sector every 12.5TB, and an enterprise level disk is rated ten times better again, at 10¹⁵. With four 16TB drives, there are approximately 48TB of data to be read without a single error from the remaining three when one fails; hence, we are looking at a sum, not an advertising slogan. On the other hand, drives in service regularly live longer than they are rated. Neither of these statements cancels out the other, and both are arguments for the exact same operating sequence: the images, followed by the rebuild, because once an image has been made, the rebuild will cost you nothing.

One job, followed all the way through.

LDN · ADR-2026-3372JOB LOGGED ✓

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

Nineteen bad sectors and that was all of it. It seemed as though two disks in the set had failed together but, in reality, one member had dropped out months ago before anybody noticed. Once another failed partway through a rebuild process, nothing was holding the volume intact. A copy of both disks demonstrated how little of that second disk could not be read, and the stripe was constructed from three good disks and a fourth that was almost good. The end-of-year filing deadline was met.

19 unreadable sectors, and no second failed disk6 days here, and back to you

What helps, and what harms.

Do this much first

Cut the power to the chassis immediately when a member drops
Label each drive with the bay in which it resides prior to pulling it
Get a quote first, then send the whole set, including rejected members
Tell us the controller model and anything that has already been tried

What sets us back

Attempting to begin rebuilding a RAID set while leaving out one member of the set
Pushing a rejected disk back into service via the card's own control panel
Allowing off-the-shelf software to run against the drives
Replacing a failed unit with a spare and allowing the card to initialise

Questions answered before you commit.

The card shows two members failed. Have we lost the volume?

Infrequently. Members are removed by controllers on their own terms (a marginal delay in a few sectors responding can cause this) and being dropped is most definitely not equivalent to being finished. No action is taken prior to all members having been imaged onto storage belonging to us; parity information from the remainder provides what one image alone may not provide. Two members flagged as bad by a controller is an average week for us. Having two members genuinely beyond imaging is uncommon.

We are degraded and one member short. Should we rebuild?

Not before images are created. In order to perform a rebuild, each remaining disk in the chassis must read itself completely end to end, which is exactly what a failing disk cannot do, and the rebuild writes new parity across the top of the existing parity, and it was that existing parity that would have reassembled your volume. Taking images first means that performing a rebuild thereafter costs you nothing, since no subsequent operations will reach an image.

Are the URE numbers worth worrying about?

A URE is one sector that the drive's own error detection/correction mechanisms cannot repair. At that point, the drive ceases reading and reports, instead of returning something that is incorrect. Consumer specifications indicate roughly one such sector per 12.5TB read; enterprise indicates one per 125TB. Those should be viewed as a minimum standard that a manufacturer will stand behind, not a countdown. Modern firmware simply flags the stripe as bad and continues.

Do the disks need the controller with them?

No. The disks themselves are sufficient. Each member maintains a record of how the set was constructed, and where that record is damaged, the shape of your data answers the same question. Construction occurs through software over the images; nothing is ever written to your original disks; and the card remains in place where it has been bolted.

Begin here if yours is doing the same thing.

RAID and server jobs → Servers that refuse to come up → Nothing detects the disk → Every device we take →

Nothing gets worse while the power is off.

It doesn't cost you anything for us to look at it. All we do is give you a breakdown of what opened and what did not, plus a single fixed price to complete, written down before you agree to let us continue. On most jobs, you only pay if we recover your data. Please don't turn on the drive until you get that breakdown.

0203 0868633

Get it assessed Sending it to Guildford

Drives, cards and entire machines

Whatever it is doing now

Footage, evidence and reports

What it costs, and what follows