Specialist work · RAID 5 arrays
A RAID 5 will carry on with one member down, and that is the whole of its promise. Most Exeter sets reach the bench with that allowance gone, a rebuild already attempted and a second drive dropped partway. The saving grace: ‘failed’ rarely means dead, and what survives is lifted from sector-by-sector clones — your originals are never written to.
Talk it over with an engineer
0800 6890668
| What's wrong | What it usually means | Step one |
|---|---|---|
| One disk gone; the set is running degraded | The cushion is used up — every hour it stays on is a gamble | Power down before drive two goes |
| The rebuild started, then fell over partway | A full-surface pass tripped on bad sectors in a second, tired disk — the classic finish | Stop; do not try again |
| Two disks shown as failed at once | Far more often one true corpse plus one drive with a scatter of bad sectors than two total losses | Post us both; imaging gives the verdict |
| Drives forced online to resurrect the set | Stale copies now sit inside newer stripes | No further steps |
| Swapped the controller; the array went missing | A metadata argument — the data underneath is untouched | Reconstruction — routine for us |
| Recovery software has been let loose on the raw disks | The originals took writes | Say what ran, then post it |
Your device is logged under its own case number the day it arrives. An engineer finds the fault, tells you plainly what stands a real chance of coming back, and writes you one fixed quote. Nothing is charged for that look, nothing obliges you, and paid work starts only when you say so.
Each drive is imaged end to end onto our storage — the two the controller wrote off included. Unstable areas get mapped, the frailest surfaces wait till the last pass, and your originals stay out of harm’s way.
Disk order, stripe size and parity rotation come from two witnesses — the metadata controllers leave on their members, and patterns in the data itself. No working card needed, nothing guessed.
Across the images we rebuild the volume: every block comes from whichever copy read it best, and parity fills in wherever no member gave a clean read — including regions only the ‘failed’ drives still carry. The file system on top is then mended and proved.
Before any fee falls due you see a full listing of everything recovered and approve it. Your files travel back on brand-new media, return postage on us, and the case stays open until you've confirmed everything opens on your own machine.
Run the sums once, then decide calmly: consumer drives are specified at one unrecoverable read error (URE) in 1014 bits read — one duff sector per 12.5TB or so — while enterprise drives are rated at 1015. Rebuilding a four-drive set of 16TB disks means pulling roughly 48TB off the three survivors, so the exposure on paper is real. In practice most drives clear their rating with room to spare. Both statements hold, and together they give the rule this page keeps repeating: get images first, then rebuild freely.
Disk two had been dead since June; no one noticed until August, when disk four dropped out partway through a rebuild. Under imaging, drive four turned out to hold nineteen bad sectors rather than a full failure. Three good members plus that near-complete fourth rebuilt the stripe, and every filing went in on time.
In most cases, yes. A controller's 'failed' often amounts to a few bad sectors, not a drive that's actually gone — and because we rebuild from images, each block can be taken from whichever copy reads best. Genuinely wrecked pairs are unusual; a pair merely kicked out of the array is what we normally see.
Only once every member has been imaged. Rebuilding reads every sector of each surviving disk — exactly the load that finishes off a marginal drive — while writing fresh parity over the very data a reconstruction would depend on. With images taken first, the identical rebuild is completely safe.
A URE is a sector too damaged for the drive's own error correction to rescue. The quoted rates are about one per 12.5TB read for consumer disks and one per 125TB for enterprise. Respect it without panicking: those figures are worst-case floors, and a modern controller sacrifices a single stripe to a URE rather than the whole array.
Yes. Controller metadata on the member disks records the geometry, and even when that's mangled the data's own patterns give it away. Every rebuild here happens from images at the bench — never from your original disks.
Each power-up of a failing device takes more data with it. Open a case first — the look-over is free whichever way it goes.