· Atomic Racks
Best Used Enterprise Drives for TrueNAS (and How to Choose Them)
Transcript
The best used drives for TrueNAS are 7200 rpm CMR enterprise models with error recovery control, bought in matched capacities for each vdev: Seagate Exos X16 and X18, and WD Ultrastar HC530 and HC550, in SATA or SAS. Avoid SMR entirely. During a ZFS resilver, a shingled drive's cache fills and write throughput collapses, so a resilver that should take a day can take a week. The Exos and Ultrastar enterprise capacity lines are all CMR. Use a host bus adapter in IT mode, not a RAID controller, which hides SMART data. LSI 9207-8i and 9300-8i cards are the common choices. SATA and SAS both work, and used SAS is often cheaper. A vdev uses its smallest member's capacity for every member, so match capacity; mixing brands is fine. RAIDZ2 widths of six to ten drives are usual: six-wide loses one third to parity, ten-wide one fifth. Desktop firmware can retry a bad sector for over a minute, which looks like a hung disk to ZFS. Enterprise drives ship with error recovery control enabled, typically at seven seconds, so the pool reads from parity instead. The full guide and our per-unit test reports are at atomicracks dot com.
The best used drives for TrueNAS are 7200 rpm CMR enterprise models with error recovery control, bought in matched capacities for each vdev: Seagate Exos X16 and X18, and WD Ultrastar DC HC530 and HC550, in SATA or SAS. Avoid SMR entirely, use an HBA in IT mode, and size RAIDZ2 vdevs at six to ten drives.
Why enterprise drives suit ZFS
ZFS was designed for datacenter storage and it assumes datacenter drives. It issues large parallel reads during scrubs and resilvers, expects drives to report errors promptly rather than retry for a minute, and treats every block checksum mismatch as a reason to read from redundancy. Enterprise drives match those assumptions by design. Desktop drives do not, and NAS-class drives sit in between.
Used enterprise drives add a price advantage. A datacenter pull at 30,000 hours sells for a fraction of a new NAS drive of the same capacity, and the platform is rated for heavier duty. The trade is that someone has to verify the drive's condition. Our post on whether used enterprise drives are safe covers the evidence.
CMR only, no SMR
Shingled magnetic recording (SMR) drives overlap tracks to increase density. Writes to a shingled zone require rewriting neighboring tracks, which the drive handles in the background using a cache area. Under sustained random writes, which is exactly what a ZFS resilver generates, the cache fills and write throughput collapses. Resilvers that should take a day can take a week, and some SMR drives drop off the bus under the load, which ZFS reads as a failed disk.
Every drive in the Exos and Ultrastar enterprise capacity lines is conventional magnetic recording (CMR). The same is true of the WD Gold and IronWolf Pro lines. If a listing does not state the recording technology, look it up on the manufacturer's datasheet before buying. For current models, both Seagate and Western Digital publish CMR and SMR designations.
SAS or SATA
Both work well under TrueNAS. The choice comes down to what controller you have and what you pay per terabyte.
| Factor | SATA | SAS |
|---|---|---|
| Connects to motherboard ports | Yes | No, needs an HBA |
| Connects to a SAS HBA | Yes | Yes |
| Interface speed | 6 Gb/s | 12 Gb/s |
| Practical difference in throughput | None for a single HDD | None for a single HDD |
| Used price per TB | Usually higher | Often lower |
| Sector format on used units | 512e | 512e, 4Kn, sometimes 520 or 528 byte |
A single hard drive cannot saturate even a 6 Gb/s link, so the interface speed is irrelevant for a disk pool. SAS drives are often cheaper used because fewer buyers can plug them in. If you are already buying an HBA, SAS opens up more inventory. Our SAS vs SATA for a homelab NAS post goes deeper, and the SAS HDD category shows what is in stock.
Use an HBA in IT mode
TrueNAS wants direct access to each disk. A RAID controller, even one configured for single-disk volumes, hides SMART data, interferes with error reporting and adds a cache layer ZFS does not know about. The standard answer is an LSI-based host bus adapter flashed to IT (initiator-target) firmware, which presents disks as plain SAS or SATA devices.
The LSI SAS 9207-8i (PCIe 3.0, 6 Gb/s) and SAS 9300-8i (PCIe 3.0, 12 Gb/s) are the common choices, each with two SFF-8643 or SFF-8087 ports driving four drives apiece through breakout cables. Both are well supported by the TrueNAS kernel. Many used cards arrive already in IT mode; check the firmware version during boot and plan to update it if it is several releases behind. The TrueNAS documentation at truenas.com maintains a hardware guide with current recommendations.
Match capacity within each vdev
A ZFS vdev uses the smallest member's capacity for every member. Put one 14 TB drive in a vdev of 18 TB drives and you get a vdev of 14 TB drives. Different vdevs in a pool can have different sizes, but within a vdev, buy matched capacity. Matched model is not required; mixing an Exos X18 with an Ultrastar HC550 of the same capacity works fine, and some people prefer it to avoid a shared firmware bug.
Two "18 TB" drives from different manufacturers usually report the same LBA count, but not always. TrueNAS tolerates small differences by leaving a little space unused. It only bites when a replacement drive is slightly smaller than the one it replaces, which ZFS will refuse.
RAIDZ2 width
RAIDZ2 tolerates two drive failures per vdev. For drives in the 14 to 22 TB range, resilvers take a day or more, and a second failure during that window is the scenario RAIDZ2 exists for. Widths of six to ten drives are the usual range. Narrower vdevs give better performance per terabyte and faster resilvers. Wider vdevs waste less space on parity. A six-wide RAIDZ2 loses one third to parity; a ten-wide loses one fifth.
For a home or small-office pool with large drives, eight-wide RAIDZ2 is a reasonable default. Beyond twelve, consider RAIDZ3 or split into two vdevs. The RAID calculator shows usable capacity and cost for each layout.
4Kn versus 512e
Modern large drives use 4,096-byte physical sectors. A 512e drive presents them to the host as 512-byte logical sectors for compatibility; a 4Kn drive presents them as 4,096-byte logical sectors. ZFS handles both correctly as long as the pool's ashift is 12, which TrueNAS sets by default. The one thing to avoid is mixing 512e and 4Kn drives in a vdev you later need to replace into, since some controllers and boot environments handle 4Kn differently. SAS drives from storage arrays sometimes arrive with 520 or 528 byte sectors and need to be reformatted before any OS will use them; see using a SAS drive in a desktop or NAS.
Error recovery control
When a drive hits a sector it cannot read, desktop firmware retries for a long time, sometimes over a minute, because it assumes it is the only copy of the data. In a ZFS pool that stall looks like a hung disk. Enterprise and NAS drives support a configurable time limit (called TLER by WD, ERC by Seagate, and exposed as SCT Error Recovery Control in smartctl) that caps the retry at a few seconds so the array can read from parity instead. Enterprise drives ship with it enabled, typically at seven seconds. You can confirm with smartctl -l scterc /dev/sdX.
FAQ
Can I mix drive brands in one vdev?
Yes. Same capacity is what matters. Mixed brands within a vdev is common and avoids correlated firmware failures.
Do I need 7200 rpm drives for TrueNAS?
Not strictly, but every current large-capacity enterprise drive is 7200 rpm, and the faster spindle shortens scrubs and resilvers. 5400-class drives are fine for cold storage.
Should I burn in used drives before adding them to a pool?
If they arrive without a recent extended self-test, yes. Run smartctl -t long and a full read pass with badblocks or dd, then check attributes 5, 197 and 198. If they arrive with a published test report showing a passed self-test and clean attributes, a short self-test on arrival is enough to confirm they survived shipping.
What we do at Atomic Racks
Every used drive we sell gets a SMART extended self-test, a fio benchmark and NIST SP 800-88 sanitization with read-back verification, and SAS drives are reformatted to 512 or 4,096 byte sectors so they work out of the box. Hours are read from the drive and never reset, and units are priced into power-on-hour tiers. The QR tamper seal on each certified unit opens its public test report. 90-day warranty on standard items, up to 2 years on certified tiers, 30-day returns and free shipping to the lower 48.
Further reading: Shucked vs enterprise drives, Building a 100TB+ homelab array on used enterprise drives.