Bench record · NAS & RAID · RDG-2025-1792
The Power Cut Cost Them Nothing. The Rebuild Did.
The mains had gone off on the Saturday, and a building-services engineering subcontractor on the Worton Grange estate in Reading walked back in after the bank holiday to find two members missing from a six-bay shelf. One of them had been amber since the Friday
. Nobody acted on it, and a second disk had gone by the Tuesday. Their IT firm put a spare in the empty bay and set a rebuild going
; that rebuild had, come Wednesday, stalled somewhere near sixty per cent
. Losing the mains, in the end, cost this firm nothing whatever. The rebuild cost it a fortnight.
Same fault on yours? Tell us what it is doing.
0800 6890668
What actually caused it.
Two failed members can be worked with. A rebuild started on top of them cannot. RAID 5 keeps one disk in reserve, and that reserve was spent the morning the first bay showed amber; once the second went, every stripe was missing two values with no way to derive either. Begin a rebuild in that state and it returns nothing while doing real harm, writing new parity through the stripes a recovery needs to find untouched. Server shelves bring their own difficulties as well. SAS members are frequently formatted to sector sizes a desktop machine cannot address, and the array's own structures sit where consumer tools will either overwrite them or walk straight past. Underneath all of that was an amber light nobody answered. Server and RAID work is published from £500 + VAT, and the assessment before it is free.
What this one needed on the bench.
How every job runs →| Bench tools | Its part in this job | Why it is here |
|---|---|---|
| PC-3000 SAS/SCSI | All six server disks read over a native SAS interface | A desktop machine cannot even address a SAS or SCSI disk from a server |
| Atola TaskForce 2 | All six imaged concurrently instead of one after another, which saved days | Takes every member of a set at once, so nothing sits waiting behind another disk |
| UFS Explorer RAID Recovery | Geometry taken from the metadata, then a virtual volume mounted over the images | Puts the array back together, then unpicks whatever volume manager the NAS laid over it |
What the bench did.
Every disk gets imaged, including the healthy four
The shelf itself was never powered up here. Members came out one at a time, each labelled with its bay, and the four healthy ones went onto imagers first, so that neither failed disk was asked to do anything before it had been assessed properly. Get the interface wrong at that point, or the sector size, and every image taken afterwards is quietly worthless.
What the two failures still held
One member was losing a head, the other was growing fresh bad sectors while being read, and neither was anything like unreadable. Each was read as a series of short passes, resting in between, and most of both surfaces ended up inside an image. Doing each of them twice paid for itself: wherever a sector differed between passes, the cleaner result was kept, since parity was in no state to arbitrate.
The metadata decides the geometry
Nothing was inferred from how that manufacturer's controllers usually behave. Member order, stripe width, parity rotation and parity delay were all taken from the structures the array had written for itself. With those four known, a virtual volume was mounted across the six images, the filesystem came out of it, and the physical members stayed powered down throughout.
What went back.
The volume came up, and everything on it was checked against the directory tree it had been carrying before a byte went onto new media, carriage home paid at this end. One qualification stands on the file: a live project folder came back short, the Wednesday rebuild having already written across that band. A copy of that folder from a fortnight earlier was still on the engineer's own laptop.
If this sounds like yours, start here.
More RAID & NAS work, written up.
If that sounds like yours, stop there.
Post it in as it is, switched off. Looking costs nothing, and you get one fixed figure in writing before any chargeable work begins — £250 + VAT for a card or stick, £300 + VAT for a single drive, from £500 + VAT for RAID and NAS. On most jobs, nothing is charged unless your data comes back.