Job log · NAS & RAID sets · ADR-2025-3192
The Machine Had Failed. The Array Had Not.
What they had was a laboratory full of people waiting and one workstation that would not come on. A genetics department, a Dell T7910, and a machine that will not boot due, it is believed, to a motherboard problem
. Behind that machine sat a 3TB RAID array carrying the group's working files, and their question was simply whether they could recover the data from the 3TB RAID array
without it. They had not touched the members and they had not let anything rebuild. Both of those decisions were worth more to this job than anything we did afterwards.
Same story on yours? Tell us what it does — the look is free.
0203 0868633
What failed, and why.
A mainboard and an array live in the same steel box and have almost nothing to do with one another. One is the machine that reads; the others are the disks that hold. When the reader stops, the holding does not, and there is no mechanism by which a fault in the mainboard reaches out and alters what is written on a platter. That is worth saying plainly, because the instinct in an office is to treat the whole tower as one broken object and to price a recovery as though the array had failed too. The genuine hazards here are the ones that come next: putting the members into a second machine to see what happens, letting a controller decide the set needs initialising, or accepting a rebuild offered by a card that has never seen these disks before. Any of those writes to members that were perfectly readable an hour earlier. Anything holding a second disk opens at £500 + VAT on the published bands, and the assessment ahead of it is free.
What the bench did with it.
The order of the job →| Equipment used | Its work on this job | The reason for it |
|---|---|---|
| PC-3000 Express | Each member profiled on its own before the set was thought about | Reads the health a controller only summarises, and does it without the array's card in the way |
| Atola TaskForce 2 | Every member copied at once behind write blockers, instead of each waiting on the last | All members of a set copy at once, so no disk sits waiting on another |
| UFS Explorer RAID Recovery | Member order and stripe geometry read off the disks, then a volume laid over the copies | Recovers the stripe order, then unpicks the volume manager the NAS layered above it |
What the bench did with it.
Order recorded, then copies before anything else
The machine arrived with its members still in their bays, which is the best start a job like this can have. Each came out in order, was labelled with the bay it belonged to, and went onto an imager. No member was asked to present a volume at that stage and none was put into another machine to see whether it would come up. A set is only ever as sound as its weakest copy, so the copying happened first and the thinking happened afterwards.
What each member had to say for itself
Health was read from the disks themselves rather than off the controller that had been summarising them. That distinction matters on an array: a card reports a member as good or failed and nothing in between, while the member itself will tell you about the sectors it has had to work at, the resets it has logged and the hours it has run. Every one of these answered properly. The department had assumed the array was fine, and for once the assumption was correct.
The geometry comes off the members
Nothing was assumed about how this controller usually arranges things. Member order, stripe width and the parity rotation were all worked out from what the disks had recorded about themselves, and a volume was then built over the copies from those numbers. The originals took no further power from that point on. Everything after it — the file system, the folders, the checking — happened on copies.
What went home.
The volume came up, the file tree was checked against it folder by folder, and the group's working files went back on fresh media. The T7910 was never repaired and never needed to be; whether it starts again is a question for somebody else entirely. Nothing was written to any member, so the array they sent is the array they still have. Work on a multi-disk box starts at £500 + VAT, and they had that in writing before a disk left its bay.
Begin here if yours is doing the same thing.
Other RAID & NAS work on the log.
Yours doing something similar? Power it down first.
Prices are published and fixed. Memory cards and USB sticks are priced at £250 + VAT. A drive on its own comes to £300 + VAT. Anything holding a second disk, whether RAID, NAS or a server, opens at £500 + VAT. The first look costs nothing, and the figure that follows is fixed and in writing before a penny is charged. Switch it off and post it in. Most jobs carry no fee where no files return.