A failed RAID array is one of the most stressful events a business can face. The good news: with the right engineers, most RAID failures are recoverable – including cases where the rebuild already failed, multiple drives are dead, or the controller is gone.
ACE RAID Recovery has recovered data from RAID arrays for over 40 years, with a 98% success rate. We work on RAID 0, 1, 5, 6, 10, 50 and 60 – from a two-disk mirror to a 28TB enterprise SAN – in our labs in Dallas, Houston, and Chicago, or securely remote via our ACE Remote Recovery Manager (ARRM).
Why RAID arrays fail
One or more disks die; with RAID 5, two active failures stop the array.
A rebuild that dies partway corrupts more data and kills the array. We see this in about 30% of RAID 5 cases that reach our lab.
The controller dies, and the array can’t be assembled without it.
Vendor firmware quietly corrupts metadata.
Arrays re-initialized, snapshots deleted, LUNs unmounted.
How a RAID array fails, step by step
Almost no array dies in a single instant. It goes through a sequence, and the decisions made at each step decide whether the data is recoverable. The first drive drops out — a bad sector, a firmware timeout, a loose backplane connection, a head crash. The controller marks it failed and, depending on the level, the array keeps serving data in degraded mode with no redundancy left.
Then the rebuild starts. The controller reads every block from the surviving disks to reconstruct the missing one. On a 12 TB array this can run for 20 hours or more of continuous full-load reads on drives that are usually the same age, from the same batch, and have the same wear. If any surviving drive has an unreadable sector in the wrong place, the rebuild hits it, the controller drops that drive too, and the array goes offline with two members missing.
That last part is the one people underestimate. A drive can pass a SMART test and still hold unreadable sectors — SMART reports what the firmware chooses to report, not whether every sector can actually be read back. A RAID 5 array with consumer drives and a 1014 uncorrectable-error rate has a mathematically real chance of a second failure during a full rebuild, and on large modern drives the rebuild window is measured in days, not hours.
The other common path is metadata. Controllers store array configuration in a DDF or vendor-specific metadata area on each member disk. If that area is overwritten — a controller replacement, a firmware update, a re-initialization, a disk moved to a different port and auto-imported — the array can look perfectly healthy to the hardware while pointing at the wrong stripe order. Writing to it in that state destroys data quickly, and it is the single most common reason a “simple” case becomes unrecoverable.
Human error completes the list: a re-initialized volume, a deleted LUN, a snapshot removed to free space, a filesystem formatted on top of a healthy array, ransomware that encrypted the array instead of a single host. None of these are hardware failures, and all of them are recoverable far more often than people expect — as long as the array is powered down and imaged before anything else is written to it.
Symptoms that mean stop, not rebuild
If any of the following is true, the array should be powered down and imaged by an engineer before another rebuild, fsck, chkdsk or “repair” is attempted:
- Two or more member disks are offline, or a second drive dropped during a rebuild
- The controller reports the array as failed, foreign, or “not initialized”
- The array was re-initialized, re-created, or the RAID configuration was deleted
- Drives were replaced, swapped between slots, or moved to a different controller
- Clicking, buzzing, or repeated spin-up/spin-down from any member disk
- A filesystem check (fsck, chkdsk, e2fsck) ran to completion and reported errors it “fixed”
- The operating system sees the volume but reports it as RAW, unformatted, or 0 bytes
- Data disappeared after a firmware update, controller replacement, or disk expansion
What we ask for before we start
RAID recovery is a reconstruction job, and the reconstruction is only as good as the information about the original layout. Send or describe the following and the evaluation is faster and more accurate:
All drives from the array, including the one that failed first. A disk that is dead to the controller still often contains the metadata that identifies the stripe order.
Dell PERC, HPE Smart Array, LSI MegaRAID, Adaptec, Areca, or a software stack such as mdadm, Storage Spaces, ZFS. Each stores metadata differently.
RAID level, stripe size, member count, parity rotation, and whether the array was ever expanded or migrated between levels.
What failed first, what was tried afterwards, and in what order. A rebuild attempt, a disk swap, or an fsck run changes the recovery plan completely.
Controllers, filesystems and platforms we work with
The array is only half the job — the data sits in a filesystem on top of it, and the volume manager in between can be the hardest part of the reconstruction.
Dell PERC H700/H710/H730/H740, HPE Smart Array P400/P410/P420/P440/P840, LSI/Broadcom MegaRAID, Adaptec, Areca, IBM ServeRAID, Dell EqualLogic and EMC arrays.
Linux mdadm (including arrays with a missing member and non-standard chunk sizes), Windows Storage Spaces and Dynamic Disks, macOS Core Storage and AppleRAID, Btrfs and ZFS pools.
NTFS, ReFS, ext3/ext4, XFS, Btrfs, ZFS, HFS+/APFS, VMFS, and vendor filesystems used by NAS appliances.
VMDK, VHDX and QCOW images stored on a RAID volume, LVM on top of mdadm, and double-virtualized setups where the array is inside a VM.
512-byte, 520-byte and 4K native sectors; 4K-aligned stripes; arrays created with a non-default stripe unit or an offset that does not match the default assumption.
BitLocker, LUKS, FileVault and self-encrypting drives — recovered from the raw image once the key material is available, without decrypting in place.
Turnaround and cost for RAID arrays
Pricing is per disk, quoted in writing after the free evaluation, and it does not change once work starts. The range depends on drive condition, the number of members, and how much reconstruction the array needs.
| Scenario | Typical turnaround | Typical cost |
|---|---|---|
| Logical damage — deleted volume, re-initialized array, filesystem corruption, healthy disks | 2–5 business days | From $1,000 per disk |
| One failed member plus a rebuild attempt, remaining disks readable | 3–10 business days | $1,000–$2,500 per disk |
| Multiple failed members, firmware or metadata damage, custom stripe layout | 5–15 business days | $1,500–$2,500 per disk |
| Physical failure — clicking, seized spindle, head crash, water or fire damage | 5–20 business days (clean room) | Quoted after evaluation |
| Emergency priority — dedicated engineers, lab time reserved | 24–48 hours | Premium tier, quoted upfront |
The evaluation itself is free and carries no obligation. If the data cannot be recovered, nothing is charged — that applies to the evaluation, the parts, and the engineering time alike.
Why the lab, and not the vendor’s rebuild
A controller rebuild is designed for uptime, not for evidence. It writes the reconstructed data back onto the surviving members, which means the first thing a failed rebuild does is overwrite the parity and metadata that a recovery needs. A lab does the opposite: every member disk is imaged first, byte for byte, onto known-good media, and all reconstruction happens on the images. The originals are never written to again.
That single difference is why arrays that a controller has already given up on are still routinely recoverable here — the stripe order, the parity rotation and the filesystem journal are all still present in the images, even when the hardware has decided the array no longer exists.
RAID levels we recover
| RAID level | Typical failure | Recovery |
|---|---|---|
| RAID 0 | One drive down = array dead | Possible from surviving drives |
| RAID 1 | Mirror broken, drive replaced | Possible from remaining disk |
| RAID 5 | Degraded, rebuild failed, second drive lost | Our most common recovery |
| RAID 6 | Two drives failed, double parity | Recovered in-lab |
| RAID 10 | Drives lost in both mirrored halves | Recovered in-lab |
| RAID 50 / 60 | Failures across nested groups | Complex – lab required |
Our recovery process
Free evaluation
An engineer assesses your array and tells you honestly if (and how) the data can be recovered.
Fixed quote
You approve a clear price before any work. Typical projects run $1,000-$2,500 per disk.
Recovery
In-lab or remote. We image every drive, rebuild the array virtually, and extract your data with proprietary tools.
Verification & return
You confirm your files; we deliver them on a new drive or via secure transfer.
Most recoveries take 3-15 days depending on complexity and data volume; emergency service runs 24-48 hours.
Why choose ACE RAID Recovery
- 40+ years of RAID recovery experience
- 98% success rate
- No data, no fee guarantee
- Support for virtually every controller and NAS platform
- Labs in Dallas, Houston, and Chicago; remote recovery anywhere
Call 844-223-7243 — engineers answer 24/7 — or start a case online.
No data, no fee
Frequently Asked Questions
Typically 3-15 days depending on complexity and data volume; emergency cases in 24-48 hours.
Yes. Controller status reflects its own metadata, not the data on the disks. We image every member and reconstruct the stripe layout independently of what the controller believes.
Yes – send every member, including the first drive that failed. A disk the controller considers dead usually still holds the metadata that identifies stripe order and parity rotation.
Usually yes, provided no new data was written afterwards. Re-initialization resets the controller metadata, but the stripe structure and filesystem are still on the disks and can be rebuilt from the data itself.
Yes – in most cases, even after a failed rebuild. Stop touching the array and call us.
Most projects range from $1,000 to $2,500 per disk, depending on RAID type, drive condition, and urgency.
Yes – Synology, QNAP, Drobo, ReadyNAS, TeraStation and more. See NAS data recovery.
Yes – encrypted connections, NDAs on request, and your data never leaves your control.