Data Recovery Case File · NAS & Network Storage · Copies Are Not Backups
Two Units, Replicating, Both Failed
His enquiry described an arrangement that looks like a belt-and-braces setup and has a specific blind spot. "I've got a couple of rack NAS units that have failed. I need to get the data recovered from at least one of them. They were replicating to each other, but both have failed. They were running RAID 10 across 4 disks each. It's not important to get both recovered, but I can provide both to increase the chances." Two units, mirrored, both gone. That is not bad luck so much as a predictable outcome, and it is worth explaining because a great many people build exactly this and believe they have a backup. Replication copies problems as faithfully as it copies data.
| Media | Two rack-mounted network storage units, each running RAID 10 across four 3TB disks and replicating to one another — both failed; either unit acceptable for recovery |
| Reported situation | Paired units configured to replicate · both units failed · RAID 10 configuration on each · recovery of either acceptable · both units available for assessment · no particular urgency |
| Fault class | Correlated failure across replicated systems — data-level faults propagated by design; array reconstruction required from whichever set proves most complete |
| Equipment used | Both units assessed and the more complete set identified · all members imaged write-blocked (Atola TaskForce 2) · arrays assembled offline from the images · members combined across units where required |
The decode: why both failed, and why offering both was right
What replication protects against, and what it does not: replication defends against a device failing. If one unit dies, the other has the data, and that is a genuine and useful protection. What it does not defend against is anything that affects the data itself, because the second unit's job is to reproduce the first faithfully and quickly. A file deleted is a file deleted on both. Corruption written is corruption replicated. A ransomware encryption pass is copied across within minutes. And a shared environmental cause — a power event, a heat problem, a bad batch of disks bought together and installed together — reaches both units because they sit in the same rack.
Why "both have failed" is therefore unsurprising: two units of similar age, running similar disks under similar load in the same room, are not independent. Disks from one purchase tend to fail within a similar window. That is why storage practice distinguishes replication from backup: a backup is separated in time — so yesterday's state survives today's mistake — and ideally separated in place. Replication is neither. It is high availability, which is a different and worthwhile thing, and it is routinely mistaken for a safety net.
Why RAID 10 helps here: the encouraging part. RAID 10 combines mirrored pairs into a striped set, so each piece of data exists twice within a unit. A failed unit is often one whose controller or appliance has failed rather than all four disks — and even where disks have failed, the set survives as long as no complete mirrored pair is lost. With four disks per unit and two units, there are eight members in total holding two logical copies of the same data.
Why offering both units was the right instinct: it genuinely increases the odds, and for a specific reason. Where one unit is missing a member and the other is missing a different one, a viable set can frequently be assembled by combining members across the two — because replication means both units hold equivalent data. Sending only the more promising unit throws that option away.
What must not happen: neither unit should be powered up to try again, and no rebuild or resynchronisation should be permitted — a rebuild reads every remaining member hard and is precisely when a second marginal disk fails. Disks should be removed and labelled by unit and bay.
On the bench
Both units were assessed rather than only the more promising one, since replication means equivalent data exists across the pair and a viable set can often be assembled by combining members across units. All members were removed, labelled by unit and by bay, and imaged write-blocked on the Atola TaskForce 2. The arrays were assembled offline from the images, where no appliance could attempt a rebuild — with mirrored pairs matched and, where a unit lacked a viable member, its counterpart drawn from the other unit. The contents were verified and delivered.
The outcome
Both units imaged, a viable set assembled offline by combining members across the pair, and the contents verified. Free assessment, one fixed written figure including VAT; where a drive has to be opened, 50% of parts and labour is payable upfront with the balance only on success — otherwise no recovery, no fee. The decode, for anyone relying on replication: it protects against a device failing and not against anything affecting the data, because the second system's job is to reproduce the first faithfully — so deletions, corruption and encryption all propagate, and shared power, heat or a disk batch reach both units in the same rack; a backup is separated in time so that yesterday survives today's mistake, and replication is not; supplying both units genuinely helps, because members can often be combined across them.
Two storage systems that mirrored each other and both failed
Send both units, not just the one that looks more promising. Because they were replicating, both hold equivalent data — so where each is missing a different member, a viable set can often be assembled by combining disks across the two, and sending only one throws that option away. Power both down now, don't let either attempt a rebuild or resynchronisation, and remove the disks labelled by unit and by bay. Then understand why this happened, because it's predictable rather than unlucky: replication protects you against a device dying and against nothing that affects the data itself, since the second system's job is to reproduce the first quickly and faithfully. Deletions, corruption and encryption all copy across. And two units of the same age, with disks from the same batch, in the same rack, are not independent of each other.
Send both — call Glasgow Data Recovery on 0141 404 0294; every member imaged write-blocked, arrays assembled offline, and members combined across units where that produces a viable set.
Request a quote online →
Our case files are drawn from genuine enquiries received by our laboratory over the past ten years, anonymised to protect client confidentiality. Each one describes the diagnostic and recovery procedure our engineers apply to that fault, using the equipment listed.