· Atomic Racks
Buying Matched Drive Lots: Firmware, Batches and RAID
For RAID and ZFS, buy drives of the same model, capacity, sector format and ideally the same firmware. Matched drives give every member the same size and the same error-recovery behavior, so the array performs evenly and replacements are simple. Mixing is possible, but each difference has a cost that depends on whether you run RAIDZ, mdadm or Unraid.
What "matched" means
A matched lot is a set of drives that share the same model number, so the same capacity, interface, sector format and mechanical platform. A tighter match also shares the same firmware version. It does not mean identical history: used drives in a matched lot can have different power-on hours and come from different serial ranges.
| Attribute | Why it matters | Must match? |
|---|---|---|
| Capacity (exact sector count) | Arrays size every member to the smallest drive | Yes, within a vdev or RAID group |
| Logical sector size (512e or 4Kn) | Hardware RAID and some pools reject mixing | Strongly preferred |
| Interface (SAS or SATA) | Hardware RAID groups usually reject mixing | Preferred |
| Firmware version | Same error recovery timing and known behavior | Preferred |
| Model and generation | Even performance, one set of quirks | Preferred |
| Power-on hours | Spreads wear, not required for function | No |
Why the same model helps
Exact capacity. Two "14 TB" drives from different vendors usually have the same sector count because most manufacturers follow the IDEMA capacity standard, but not every drive does, particularly OEM and array-formatted models. A replacement that is a few sectors smaller cannot replace a member of an existing vdev or md array. Buying one model removes the question.
Even performance. A RAIDZ or RAID 6 stripe completes when its slowest member finishes. One slow drive, or one drive with different caching behavior, sets the pace for the whole array during scrubs and resilvers.
Firmware behavior. Enterprise drives limit how long they retry a bad sector before reporting an error, so the array can repair it from parity instead of timing the drive out. Matching firmware means every drive handles errors the same way, and if a firmware update is ever needed you follow one procedure for the whole set. Check the version with smartctl -i. OEM firmware from Dell, HPE or NetApp can differ from the generic version of the same model.
Mixing capacities and vendors
- ZFS RAIDZ. Each vdev uses the smallest member's capacity on every drive. You can mix vendors of the same size without trouble. A vdev grows only after every member is replaced with a larger drive and autoexpand is on. Different vdevs in one pool can have different sizes.
- Linux mdadm. The array uses the smallest component size, like ZFS. Partitioning every drive slightly below full size protects you against a future replacement being a few sectors short.
- Unraid. Built for mixed capacities. Data drives can be any size; the parity drive must be at least as large as the largest data drive. See the Unraid parity drive guide.
- Hardware RAID controllers. The strictest. Many refuse to mix SAS and SATA, or 512 and 4Kn, within one virtual disk.
Mixing 512e and 4Kn
Both formats use 4K physical sectors; they differ in the logical size presented to the host. ZFS handles mixing as long as the pool was created with ashift=12 (4K alignment), which modern defaults choose. A pool created with ashift=9 cannot accept a 4Kn drive. In mdadm, the array adopts the largest logical sector size of its members, which can cause problems for a file system created with smaller blocks. Hardware RAID controllers usually refuse to mix. The simplest rule: pick one format for the array and buy replacements in it.
Mixing SAS and SATA on an expander
A SAS HBA or expander backplane can run SAS and SATA drives at the same time, with SATA traffic tunneled through the expander. ZFS and Unraid do not care which is which. Two cautions apply. SATA drives are single-ported, so dual-path enclosures cannot use the second path for them. And a misbehaving SATA drive on some older expanders can cause resets that affect other drives on the same expander. Keeping one interface per array avoids both. Our SAS vs SATA guide covers the trade-offs.
Why some people deliberately mix batches
There is an argument against perfect matching. Drives built in the same production run, installed together and worked identically may wear out at similar times, which raises the chance that a second drive fails during a rebuild. Some builders spread purchases over time or buy from different batches for that reason.
With used enterprise drives the trade-off looks different. A lot of one model will usually include different serial ranges and a spread of power-on hours, which gives you model consistency with some natural variation. If you want a deliberate spread, buy the same model across two tiers or two purchases. Double parity, a spare and backups handle correlated risk better than mixing models does.
How our lots are built
We sell lots by model, so a lot of Seagate Exos X16 14 TB (ST14000NM001G) drives is all that part number, and a lot of WD Ultrastar HC530 14 TB SAS (WUH721414AL5204) drives is all that part number. Power-on hours are read from each drive and published on its test report, and drives are grouped into usage tiers by hours, so you know the wear on every member before you build. If you need a specific firmware version, sector format or a mix of tiers, ask in a bulk quote request. The HC530 vs HC550 vs HC570 comparison helps if you are choosing between generations, and you can browse 14 TB drives.
Warm spare math
A spare on the shelf shortens the window between a failure and the start of a rebuild, which is when an array is most exposed. To size it, multiply drive count by an annualized failure rate. As an assumption, not a measured figure for any model, use 1 to 2 percent per year:
| Array size | Expected failures per year (1 to 2% assumed) | Over 5 years | Suggested spares |
|---|---|---|---|
| 6 drives | 0.06 to 0.12 | 0.3 to 0.6 | 1 cold spare |
| 12 drives | 0.12 to 0.24 | 0.6 to 1.2 | 1 cold spare, or 1 hot spare if unattended |
| 24 drives | 0.24 to 0.48 | 1.2 to 2.4 | 1 hot spare plus 1 cold spare |
Expected failures are an average, not a promise; a single array can see none or several. The spare should be the same model, tested before it goes on the shelf, so you know it works when you need it. A rebuild of a full 14 TB drive at an average of 150 to 200 MB/s takes roughly 20 to 26 hours, which is why double parity is the norm at these sizes.
FAQ
Can I mix Seagate and WD drives in one ZFS vdev?
Yes, if the capacities match. The vdev runs at the pace of its slowest member.
Do drives in a RAID need identical firmware?
It is not required for function, but matching firmware gives consistent error handling and one update procedure.
Should all drives have similar hours?
Not necessarily. A spread is fine and may reduce the chance of drives wearing out together.
Is a hot spare better than a cold spare?
A hot spare starts rebuilding immediately but runs and wears in the meantime. A cold spare suits arrays you check regularly.
What we do at Atomic Racks
Every drive we sell passes a SMART extended self-test, a throughput benchmark and NIST SP 800-88 sanitization with read-back verification, as described on how we test. Each drive carries a QR tamper seal linking to its public test report and a tier-based warranty (see warranty). Shipping is free within the US from Englewood, Colorado, with 30-day returns (buyer-paid return postage unless the item is defective).