In this guide: 13 sections
- Which drive to buy, by bay count and workload
- What to check on the listing, and on the day the drive arrives
- Error recovery time is the one specification that changes behavior
- Workload rating is a warranty condition, and scrubs spend most of it
- Bay ratings and vibration sensors measure less than their marketing implies
- Reliability figures, and what each one is actually measuring
- The 10^14 argument, done with the correct probability
- Rebuild time, and drive count against drive size
- RAID levels and filesystems: what each one survives and gives up
- SMR, shucked drives and mixed batches
- Helium, power, noise and heat
- Burn-in, with the commands and the badblocks ceiling
- Capacity planning, and the cost of expanding later
Buy CMR NAS drives: WD Red Plus or Seagate IronWolf for up to eight bays, Red Pro, IronWolf Pro or enterprise drives above that, and never a shingled (SMR) drive like the plain WD Red at 2 to 6 TB with EFAX in the part number. The Pro and enterprise tiers are also the answer at any bay count when the array is busy around the clock, because they are rated for 550 TB of traffic a year against 180. A NAS drive and a desktop drive of the same generation often share platters, heads and firmware base; the difference that actually breaks arrays is how long the drive keeps retrying a bad sector before it answers, so check that setting with smartctl -l scterc on the day each drive arrives. After that, read the datasheet for the exact part number, because workload rating, error rate and fill gas all change inside a single family.
Which drive to buy, by bay count and workload
The bay counts are the vendors’ own. Western Digital’s Red Plus datasheet rates the family for “up to 8 bays” and 180 TB of traffic a year. Seagate’s 2022 IronWolf datasheet lists “1-8 Bays” and a workload rate limit of 180 on every capacity from 1 TB to 12 TB, and states that “All IronWolf drives utilize Conventional Magnetic Recording (CMR)”. The Pro lines have dropped the ceiling. Western Digital’s January 2025 WD Red Pro datasheet rates the whole family, 2 TB through 26 TB, at 550 TB per year and calls it suitable for “RAID-optimized NAS systems with unlimited # of bays”. Seagate’s current IronWolf Pro datasheets rate its equivalent identically - 550 TB per year, “Drive Bays Supported: Unlimited”. Enterprise nearline drives carry the same 550.
| Your NAS | Buy | Why |
|---|---|---|
| 1 to 8 bays, home or small office | WD Red Plus, Seagate IronWolf (such as ST8000VN004 or ST12000VN0008) or Toshiba N300, from the NAS-class listings | 180 TB/year (WD Red Plus, IronWolf, Toshiba N300); Red Plus “up to 8 bays”, IronWolf “1-8 Bays” |
| More than 8 bays, or busy around the clock | WD Red Pro, IronWolf Pro, or an enterprise nearline drive such as Exos, Toshiba MG or Ultrastar | 550 TB/year; both Pro lines rated for unlimited bays |
| Used, to save money | Enterprise pulls, SATA or SAS, priced per terabyte against new NAS drives | Check hours, sector format and pin 3 first: see enterprise drives |
| Never in an array | Any drive-managed SMR drive, including the plain WD Red at 2, 3, 4 and 6 TB with EFAX in the part number, such as WD20EFAX and WD40EFAX | Stalls for minutes during a rebuild, and no timeout setting rescues it |
The ladder that matters runs inside a family, not between the two vendors. WD Red Plus is 180 TB/year against Red Pro’s 550. Two WD Red Pro part numbers of the same 14 TB capacity differ by a factor of ten on unrecoverable read error rate. Two WD Red Plus part numbers of the same 12 TB capacity differ by more than two to one on idle power, because one is helium-filled and one is not. The part number is the specification. The family name is a shelf.
The bay ratings themselves are marketing tiers more than measurements (the vibration section explains why), and how many drives to buy at what size is a trade that the rebuild section works through.
What to check on the listing, and on the day the drive arrives
- Filter to CMR first, and treat “not stated” as SMR. Start at the hard drive listings and keep the drives whose listing says CMR, or hide the ones whose listing says SMR. Both read the seller’s words, and most listings say neither, so check the exact part number against the maker’s own CMR and SMR list before you buy. This is the one check where a wrong guess costs you an array rather than an afternoon.
- Narrow to the class you need, not the one with the best badge. NAS-class and enterprise-class listings are separate filters here. Then read the datasheet for the exact part number on the URE rate, the workload figure and the fill gas - all three vary inside a single family.
- Check the price step between capacities before deciding drive count. Sort by price per terabyte, or set 12 TB against 20 TB. Fewer bigger drives cost more parity and a longer resilver; more smaller drives cost bays, power and spin-up surge.
- Run
smartctl -l sctercon every drive the day it arrives, used or new, before it joins anything. If the answer isDisabled, set it -70,70undermdadmor a hardware controller,1,1under ZFS if the drive accepts it and70,70if it does not - and try the persistent formscterc,70,70,pbefore writing a boot script. Read the value back every time. If the answer is “not supported”, raise the kernel timeout to 180 seconds or put that drive somewhere else. The error recovery section has the commands. - Burn it in with the right block size, and diff the SMART output.
badblocks -b 4096 -wsvup to 16 TiB,-b 8192above. Attributes 5, 197 and 198 must be zero before and after. Budget eight to ten days per large drive (burn-in). - For used drives, do all of the above inside the return window. The used-drive guide covers what the drive itself records and how to read the hours; the enterprise pulls guide covers the HBA, the sector-size trap and why a SAS drive sidesteps the error-recovery argument entirely.
- Buy deliberately mixed. Two vendors, two retailers, or the same model months apart, date codes checked on arrival. It is cheap insurance against the one failure mode parity does not survive.
Error recovery time is the one specification that changes behavior
A drive that cannot read a sector does not give up. It repositions the head, re-reads, applies progressively heavier correction, tries deliberate off-track offsets in both directions, and re-reads again. On a desktop that is exactly right: the drive is the only copy of the file. The Linux kernel RAID wiki puts the desktop-drive retry ceiling at over two minutes. No vendor publishes the retry algorithm, so treat two minutes as an observed order of magnitude, not a specification.
Now put that drive in an array. The controller asked for a sector, has had no answer, concludes the drive is dead, and drops it. You are now rebuilding - a full-surface read of every surviving drive, for days at current capacities - over one slow sector that was about to be delivered.
| Layer | Timeout | Where it is set |
|---|---|---|
| Hardware RAID controllers | Commonly around 8 seconds | Controller firmware, often undocumented |
| Linux SCSI/libata command timeout | 30 seconds | /sys/block/sdX/device/timeout |
| ZFS on Linux | Inherits the 30-second kernel timeout | As above |
| Drive with ERC enabled | 7 seconds typical factory default | SCT Error Recovery Control |
| Drive with ERC disabled | Over 120 seconds | Not settable |
| Drive-managed SMR, error path | Over 10 minutes | Not settable, not bounded |
The kernel RAID wiki names the consequence: when the kernel gives up first, “the raid code assumes the drive is dead and kicks it from the array”. The drive was working. The timeout killed it.
Make the drive give up first
The ATA feature is SCT Error Recovery Control; vendors rename it - Western Digital TLER (time-limited error recovery), Seagate ERC (error recovery control), Samsung and the old HGST lines CCTL (command completion time limit). They are the same SCT command set underneath.
$ sudo smartctl -l scterc /dev/sda
SCT Error Recovery Control:
Read: 70 (7.0 seconds)
Write: 70 (7.0 seconds)
$ sudo smartctl -l scterc,70,70 /dev/sdb # read limit, write limit
The units are deciseconds, so 70 is seven seconds - deliberately just under
the eight-second controller timeout. The smartctl manual page adds the floor most
guides omit: 0 disables the feature entirely, and “other values less than 65 are
probably not supported”. Read: Disabled means the drive has the feature but is
sitting in desktop behavior; SCT Error Recovery Control command not supported
means it lacks it.
The setting traditionally did not survive a power cycle. From smartmontools 7.3, the manual page documents a persistent form:
# set, persistent across power-on reset
$ sudo smartctl -l scterc,70,70,p /dev/sdb
# read back the persistent values
$ sudo smartctl -l scterc,p /dev/sdb
# restore the manufacturer's defaults
$ sudo smartctl -l scterc,reset /dev/sdb
The ,p and ,reset options “require the device to support ATA ACS-4 or higher”,
so an older drive will refuse them. Try the persistent form first and verify it
survives a cold power cycle, not a reboot - a warm reboot may not drop the drive’s
power rail, which makes a non-persistent setting look persistent. Where ,p is
unavailable, use a udev rule or a systemd unit that runs on every boot, and read the
value back afterwards rather than assuming the write took.
ZFS wants a far shorter limit than RAID does
The seven-second convention comes from hardware RAID controllers. The OpenZFS hardware guidance recommends something an order of magnitude tighter:
it is advisable to write a script to set the error recovery time to a low value, such as 0.1 seconds until ZFS is modified to control it. This must be done on every boot.
That is smartctl -l scterc,1,1. ZFS has a checksum, knows the block is wrong and
can reconstruct it from parity or a mirror at once, so every second the drive spends
retrying is wasted. Expect most drives to refuse it, because values below 65 are
probably not supported. Set it, read it straight back with smartctl -l scterc,
and fall back to 70,70 if the drive did not take the value.
| Stack | Recommended ERC | Why |
|---|---|---|
| Hardware RAID controller | 70,70 (7.0 s) | Just under the controller’s own ~8 s timeout |
Linux mdadm |
70,70 (7.0 s) | Named as typical in the smartctl manual page |
| OpenZFS | 1,1 (0.1 s) if the drive accepts it, else 70,70 | Parity reconstruction is faster than drive retry |
| SAS, any stack | Mode page 0x01 | Ships at array-appropriate values |
When the drive cannot do it at all
Move the other timeout. The kernel RAID wiki gives the fallback directly:
# raise the kernel's per-device command timeout past the drive's retry ceiling
for d in /sys/block/sd*/device/timeout; do echo 180 > "$d"; done
This must also be re-applied at boot. It is the worse fix - a genuinely dead drive now hangs the array for three minutes instead of thirty seconds - but it beats losing a member. Software RAID’s timeout is a kernel setting you own, and a hardware controller’s is not, so behind a hardware controller a drive that cannot lower its own timeout has no fix at all. That is why hardware RAID and desktop drives pair so badly.
Which drives allow it is a per-model fact, not a per-brand one. The NAS and
enterprise lines - WD Red Plus, Red Pro, Gold and Ultrastar; Seagate IronWolf,
IronWolf Pro and Exos; Toshiba N300 and MG - ship with it enabled or settable. Consumer lines vary.
WD removed the setting from its desktop drives and warned that the old
WDTLER.EXE utility could brick newer firmware, so never run it. Check the
drive in front of you, not the line it is in. A SAS drive keeps its
limit in the SCSI Read-Write Error Recovery mode page (0x01), usually at an
array-appropriate value, and sdparm --get=RTL reads it.
SMR is not a slow sector, it is an unbounded one
The same kernel wiki gives a worse figure for shingled media: an SMR drive can stall for over ten minutes during a read error. That is beyond any achievable kernel timeout and beyond the 180-second workaround, so ERC tuning does not rescue a shingled drive. The SMR section covers what that does to a rebuild.
Workload rating is a warranty condition, and scrubs spend most of it
Workload rating is published in terabytes per year. Western Digital’s Red Plus datasheet defines it as “the amount of user data transferred to or from the hard drive”, and the Red Pro footnote gives the formula:
Annualized Workload Rate = TB transferred x (8760 / recorded power-on hours)
Host user data only. The drive’s own background media scan and read-after-write verification do not appear in the vendor definition.
| Tier | Workload rating | Rated power-on hours | Source |
|---|---|---|---|
| Desktop | ~55 TB/year | ~2400 hours/year (light-duty rating) | Consumer datasheets, where they state it at all |
| NAS | 180 TB/year | 8760 hours/year | WD Red Plus; Toshiba N300 |
| NAS Pro | 550 TB/year, both vendors | 8760 hours/year | WD Red Pro and Seagate IronWolf Pro datasheets |
| Enterprise nearline | 550 TB/year | 8760 hours/year | Seagate Exos X24 product manual; Toshiba MG11 datasheet |
The step that is real is the one from 180 to 550, between Red Plus and Red Pro inside a single vendor’s range: a factor of three on the warranty condition. The NAS Pro tier and the enterprise nearline tier have the same published workload rating, so the workload figure is not what you are paying the enterprise premium for - see enterprise drives for what is.
Two numbers on one datasheet, quoted at two operating points
WD quotes its reliability figure at a lighter workload than its workload rating. The Red Pro MTBF footnote estimates MTBF at a workload of 220 TB/year and 40 °C, and says “Derating of MTBF will occur above these parameters, up to 550TB writes per year”.
WD Red Pro
headline MTBF 2,500,000 hours (14-26 TB) / 2,000,000 hours (2-12 TB)
quoted at 220 TB/year, 40 °C
workload rating 550 TB/year
between them MTBF derates, curve not published
WD Red Plus
headline MTBF 1,000,000 hours
quoted at 90 TB/year, 40 °C
workload rating 180 TB/year
between them MTBF derates, up to 65 °C drive temperature
So the two figures do not contradict each other. 550 TB/year on Red Pro, or 180 on Red Plus, is the most traffic the warranty covers; 220, or 90, is the traffic at which the headline MTBF was modeled. A drive run at its full rated workload does not have the MTBF on the front of its datasheet, and WD publishes only that it is lower. Design for 220 TB/year on Red Pro and 90 TB/year on Red Plus if you want the advertised MTBF.
It is still a warranty condition, not a wear counter: hard drives wear with seeking, flying hours and thermal cycling far more than with bytes moved, unlike the flash in SSD endurance. The hours line binds sooner anyway: a desktop drive rated for 2400 hours a year, run 24x7, spends its annual allowance in 100 days.
Household traffic barely touches it
55 TB/year / 31,536,000 s = 1.7 MB/s sustained, continuously
180 TB/year / 31,536,000 s = 5.7 MB/s sustained, continuously
220 TB/year / 31,536,000 s = 7.0 MB/s sustained, continuously
550 TB/year / 31,536,000 s = 17.4 MB/s sustained, continuously
saturated 1 GbE (125 MB/s) = 3,942 TB/year
saturated 2.5 GbE (312 MB/s) = 9,839 TB/year
saturated 10 GbE (1250 MB/s) = 39,420 TB/year
A saturated gigabit link is about twenty-two times a 180 TB/year rating and seven times a 550 TB/year one. A family writing two terabytes of photos and backups a year sits at roughly one percent of a NAS drive’s budget. What keeps a home NAS busy is not the humans.
Scrubbing is what spends it, and mdadm far more than ZFS
Take six 12 TB drives in RAID6, 48 TB usable, 30 TB in use. ZFS scrubs allocated
blocks only, so each drive reads roughly its own share of what is stored, about
7.5 TB per scrub. mdadm’s check is allocation-blind: it reads every sector of
every member, 12 TB per drive per check, because it compares raw parity against raw
data with no idea what the filesystem above it considers used.
And the schedule is set for you. Debian ships mdadm’s checkarray on a cron
entry that fires on the first Sunday of the month at 00:57. Red Hat and Fedora
ship raid-check, driven by raid-check.timer, whose unit file reads
OnCalendar=Sun *-*-* 01:00:00 and describes itself as a “Weekly RAID setup health
check”. TrueNAS schedules a monthly ZFS scrub per pool by default.
Debian checkarray first Sunday of the month, 00:57 -> 144 TB/yr per 12 TB member
Red Hat raid-check every Sunday, 01:00 -> 624 TB/yr per 12 TB member
TrueNAS ZFS scrub monthly per pool -> 90 TB/yr per member at 62% full
A weekly mdadm check on 12 TB drives is 624 TB/year of scrub traffic alone -
more than a 550 TB/year enterprise rating, before the array has served a single
file - and on Red Hat or Fedora it is the shipped default. Monthly is 144 TB/year,
which already consumes 80 percent of a 180 TB/year WD Red Plus allowance and 65
percent of the 220 TB/year point where WD quotes Red Pro’s MTBF.
Do not turn scrubbing off: it is how latent sector errors are found while
redundancy still exists to repair them, and
the URE section has the
evidence. Monthly is defensible on any of the three. Weekly on large mdadm
members is not. systemctl edit raid-check.timer and an OnCalendar=Sun *-*-01..07 01:00:00 is the monthly equivalent.
Bay ratings and vibration sensors measure less than their marketing implies
Vendors used to rate NAS drives by bay count, and the tiering has come apart. Only WD Red Plus and Seagate IronWolf still carry a number. Both vendors have dropped the ceiling on their Pro lines, and neither has published the test behind either the old number or the new one. A limit that can go from 24 bays to unlimited without a matching change in the mechanical specification was a marketing tier rather than a measurement.
The mechanism is real, though. The head’s position error budget while writing is, by this site’s estimate from typical platter geometry, in single-digit nanometers, and rotational vibration - angular acceleration about an axis parallel to the spindle - acts directly on the actuator arm, where linear vibration does not. Every seek pushes a reaction torque into the chassis and on to every drive bolted to the same metal.
Two accelerometers on the drive’s circuit board each measure linear acceleration; subtract one from the other and whole-drive shaking cancels, leaving a signal proportional to rotation about the spindle axis. That signal is fed forward into the servo, so the correction starts before the head has moved. A drive without the pair finds out from its position error signal alone, by which point it must inhibit the write and wait a full revolution. ASHRAE Technical Committee 9.9, in a 2019 paper whose authors include Seagate, Dell, IBM and Intel engineers, puts it in the vendors’ own words: the system “must wait for at least one revolution of the disk to attempt again to write”.
7200 rpm -> 60 / 7200 = 8.33 ms per revolution
5400 rpm -> 60 / 5400 = 11.11 ms per revolution
theoretical average access, 7200 rpm:
rotational latency 4.17 ms + seek 4 to 9 ms = 8 to 13 ms
empirical average access, 7200 rpm (OpenZFS) = 13 to 16 ms
rated average access, 15k rpm (OpenZFS) = 3.4 ms read, 3.9 ms write
Against OpenZFS’s measured 13 to 16 ms average access for 7200 rpm drives, a lost 8.33 ms revolution is a 52 to 64 percent increase on that operation. And nothing reports it. No SMART attribute increments and no error is logged; the array is simply slower than it was on the bench.
What the sensors cannot see
The ASHRAE paper is plain that the two sensors “only measure in-plane motion”:
| Disturbance | Sensed? |
|---|---|
| In-plane linear motion (x, y) | Yes, and canceled as common mode |
| Rotation about z (the spindle axis) | Yes - this is the whole feature |
| Rotation about x or y (chassis rocking) | No |
| Linear motion along z | No |
| Non-rigid-body motion (panel flexing, resonance) | Not accurately |
An “RV sensor” tick on a datasheet buys rejection of one of five disturbance modes. There is also a bandwidth ceiling. ASHRAE: “The current sets of technologies and mechanics allow for error rejection below 2 kHz, but disturbances from AMDs can input excitations up to 20 kHz”, where AMD means air-moving device, a fan or blower. The servo rejects below 2 kHz and the drive is sensitive to 20 kHz.
In a dense modern chassis the fans have overtaken neighboring drives as the main disturbance, and acoustic power goes as the fifth power of fan speed. That is why a NAS that was fine in winter can lose throughput in a warm room, with nothing in the monitoring to say so. Helium makes it worse: ASHRAE finds helium designs “more susceptible to external vibration, especially vibration caused by coupling with acoustic sources”.
The datasheet figure will not predict any of this. Seagate’s IronWolf Pro datasheet publishes rotational vibration as 12.5 at 10 to 1500 Hz, which excludes the sub-10 Hz region where chassis rocking lives, and the value is identical at 4, 6, 8, 10, 12, 14, 16, 18 and 20 TB - a design target the family is qualified against, not a per-model measurement. ASHRAE concludes that “there is no specific reference speed for an AMD that may be used to determine the HDD performance in an enclosure”.
What you can do is bound it: measure sustained sequential throughput with the fans pinned to minimum, then to maximum. If the numbers differ materially, try rubber-grommet mounting, a slower larger fan, or damping the panel. In a four-bay plastic enclosure the effect is usually small enough to ignore. By eight bays in a metal chassis it is real, and it arrives as slowness rather than as an error.
Reliability figures, and what each one is actually measuring
Vendors publish MTBF; buyers want a failure rate. They are not the same thing. WD Red Pro from 14 TB and IronWolf Pro from 12 TB are rated at 2,500,000 hours, which is 285 years and which nobody claims is a life expectancy: it is the inverse of a modeled constant hazard rate during the flat part of the bathtub curve. An AFR obtained by dividing 8,760 by the MTBF is not a published AFR; it is the same model figure in different units. Toshiba prints both for its MG11, and they agree to the digit: 8,760 / 2,500,000 = 0.35 percent.
Field data disagrees with the model. Schroeder and Gibson’s FAST 2007 study of about 100,000 disks found field replacement rates of 2 to 4 percent typical, against datasheet MTTFs of 1.0 to 1.5 million hours - a nominal AFR of at most 0.88 percent - and replacements showed autocorrelation significant at lags up to 30 weeks. Failures arrive in clusters. Backblaze’s Q1 2026 Drive Stats and its 2025 annual report give the current picture:
drives in service 341,263
drive-days in quarter 30,203,180
failures in quarter 1,030
quarterly AFR 1.24%
lifetime AFR 1.39% (529,968,464 drive-days, 20,212 failures)
20 TB+ cohort AFR 0.85% across the quarter
2025 full year 1.36% (down from 1.55% in 2024)
worst single model 4.90% (1,251 drives of one Seagate 14 TB model)
The fleet-wide figure and the worst-model figure differ by a factor of four, so a fleet average is not a prediction about a specific model. And Backblaze names its own blind spot: “Have we been completely under-reporting day one failures? Short answer: yes, probably.” A drive that arrives dead, or fails in its first week, never enters the dataset. Do not read a 0.85 percent AFR as your probability of a bad drive out of the box; read it as your probability of losing a drive that survived its first week in a datacenter. That class of failure is what burn-in exists to catch.
Google’s FAST 2007 disk failure study found a drive 39 times more likely to fail within 60 days of its first scan error, yet a large fraction of failed drives showed no SMART errors at all. When SMART says something, listen; a clean report is not evidence of health.
The 10^14 argument, done with the correct probability
Consumer datasheets specify an unrecoverable read error rate of less than 1 sector per 10^14 bits read. NAS and enterprise drives are usually 10^15, and some SAS drives 10^16. The tiers are not as clean as the marketing: the entire WD Red Plus family, 2 TB through 12 TB, is specified at less than 1 in 10^14, the same as a desktop drive, and within WD Red Pro the 14 TB WD141KFGX is specified at 10^14 while WD142KFGX, also 14 TB, is specified at 10^15.
The version of this argument you will meet online divides bits read by the URE rate and calls the result a probability:
16 TB read x 8 = 1.28e14 bits
1.28e14 x 1e-14 = 1.28 -> "128% chance of a URE"
A probability of 128 percent should have ended the argument on the spot. The division gives the expected number of UREs, not the probability of at least one. The correct form is Poisson:
P(at least one URE) = 1 - e^(-N x p)
N = bits read, p = per-bit error rate
16 TB: 1 - e^(-1.28e14 x 1e-14) = 1 - e^(-1.28) = 72.2%
72.2 percent, not 128 percent. Running the same arithmetic per 4096-byte sector instead of per bit gives the same answer to three significant figures.
The table, extended to current capacities and to 10^16
Five-drive RAID5, rebuild reads four drives in full. These figures, like the other derived tables in this guide, are calculations from the published specification ceiling, not measurements:
| Drive size | Bytes read | At 10^-14 | At 10^-15 | At 10^-16 |
|---|---|---|---|---|
| 2 TB | 8 TB | 47.3% | 6.2% | 0.6% |
| 4 TB | 16 TB | 72.2% | 12.0% | 1.3% |
| 8 TB | 32 TB | 92.3% | 22.6% | 2.5% |
| 12 TB | 48 TB | 97.9% | 31.9% | 3.8% |
| 16 TB | 64 TB | 99.4% | 40.1% | 5.0% |
| 20 TB | 80 TB | 99.8% | 47.3% | 6.2% |
| 24 TB | 96 TB | 99.95% | 53.6% | 7.4% |
| 30 TB | 120 TB | 99.99% | 61.7% | 9.2% |
Read down the columns rather than across the rows. The specification tier dominates the capacity. A 24 TB drive at 10^16 is safer on this metric than a 2 TB drive at 10^14 by a factor of six, while reading twelve times as much data. Capacity growth is not what broke RAID 5; it is what made the difference between tiers unignorable.
Why those percentages overstate the risk
- The spec is a ceiling, not a mean. Datasheets say “less than 1 in 10^14”, and no vendor publishes the distribution.
- The empirical test fails. If 72 percent per rebuild were real, five-drive 4 TB RAID5 arrays would almost never complete one. They routinely do.
- Errors are not independent. Bairavasundaram and colleagues at NetApp studied latent sector errors across 1.53 million drives in over 50,000 arrays over 32 months and found them in 8.5 percent of consumer-class disks against 1.9 percent of enterprise-class. Errors concentrate in a few bad drives, and over 60 percent of latent sector errors were found by scrubbing.
- A URE is not automatically an array loss. That is the biggest error in the whole argument:
| Stack | What a URE during rebuild does |
|---|---|
| Older hardware RAID controller | Aborts the rebuild and drops the array |
Linux mdadm |
Records the block in its bad block list and continues |
| OpenZFS | Names the affected file in zpool status -v and continues |
One loses the array; one loses nothing; one loses a named file you can restore from backup. Robin Harris’s 2007 ZDNet article “Why RAID 5 stops working in 2009” was right about direction - RAID 6 and raidz2 became the default - and wrong about magnitude, which is why arrays have not stopped working.
The probability that actually governs array loss
On a modern software RAID, the event that loses the array is a second whole-drive failure during the rebuild window. For a five-drive array with one member lost, derived from Backblaze’s published AFR and assuming independence:
| Scenario | Rebuild window | P(second failure) |
|---|---|---|
| 12 TB, 100 MB/s, AFR 0.85% | 33.3 h | 0.016% |
| 20 TB, 100 MB/s, AFR 0.85% | 55.6 h | 0.027% |
| 24 TB, 100 MB/s, AFR 0.85% | 66.7 h | 0.032% |
| 20 TB, 50 MB/s, AFR 0.85% | 111 h | 0.054% |
| 20 TB, 100 MB/s, AFR 1.39% (fleet lifetime) | 55.6 h | 0.044% |
| 20 TB, 50 MB/s, AFR 3% (Schroeder field range) | 111 h | 0.19% |
Under independence these are small numbers, which is exactly why the correlation evidence matters more than the URE arithmetic. Array members share a model, a firmware revision, a manufacturing batch, a thermal history and a workload. The exact probability cannot be derived from a published upper bound and a violated independence assumption, but RAID5’s margin gets thinner with every capacity generation - a claim about direction, which does not need the percentage to be right.
Rebuild time, and drive count against drive size
Rebuild time is capacity divided by sustained throughput. Current 3.5-inch 7200 rpm drives do roughly 250 to 287 MB/s on their outermost tracks and 110 to 130 MB/s on their innermost, and a whole-surface average near 200 MB/s is what a rebuild gets on an idle array. Within one family the rate moves a lot: WD’s Red Pro datasheet runs from 164 MB/s at 2 TB to 287 MB/s at 24-26 TB.
| Drive | At 200 MB/s (idle) | At 100 MB/s (in service) | At 50 MB/s (busy) |
|---|---|---|---|
| 4 TB | 5.6 h | 11.1 h | 22.2 h |
| 8 TB | 11.1 h | 22.2 h | 44.4 h |
| 12 TB | 16.7 h | 33.3 h | 66.7 h |
| 16 TB | 22.2 h | 44.4 h | 88.9 h |
| 20 TB | 27.8 h | 55.6 h | 4.6 days |
| 24 TB | 33.3 h | 66.7 h | 5.6 days |
| 30 TB | 41.7 h | 83.3 h | 6.9 days |
It is worse than it was because capacity has grown far faster than the rate at which a head can read a track: in 2007 a 1 TB drive at 80 MB/s rebuilt in 3.5 hours.
ZFS and btrfs rebuild only allocated blocks, so a half-full pool resilvers in
roughly half the time, a genuine advantage over mdadm, which copies every block.
Against that, a raidz resilver walks the pool in block-pointer order rather than
LBA order, turning a sequential read into a near-random one; OpenZFS’s sequential
resilver covers mirrors and dRAID, not raidz.
Fewer bigger drives cost more parity, not less
“Fewer bigger drives” is the standard advice and it is half right. Here is the whole trade, at a fixed 60 TB usable on raidz2 or RAID6, with resilver times at 100 MB/s:
| Layout | Raw | Parity tax | Per-drive resilver | Data read during resilver |
|---|---|---|---|---|
| 8 x 10 TB | 80 TB | 25.0% | 27.8 h | 70 TB |
| 7 x 12 TB | 84 TB | 28.6% | 33.3 h | 72 TB |
| 6 x 15 TB | 90 TB | 33.3% | 41.7 h | 75 TB |
| 5 x 20 TB | 100 TB | 40.0% | 55.6 h | 80 TB |
| 4 x 30 TB | 120 TB | 50.0% | 83.3 h | 90 TB |
At a fixed parity count, a narrower vdev spends a larger fraction of its raw capacity on redundancy, and from eight drives to four the resilver is three times longer and reads 29 percent more data. Against that, drive count multiplies the annual chance of touching a rebuild at all. At Backblaze’s 20 TB+ AFR of 0.85 percent, and at their lifetime 1.39 percent:
| Drives | P(at least one fails in a year) at 0.85% | at 1.39% |
|---|---|---|
| 4 | 3.36% | 5.45% |
| 5 | 4.18% | 6.76% |
| 6 | 4.99% | 8.06% |
| 7 | 5.80% | 9.33% |
| 8 | 6.60% | 10.59% |
| 12 | 9.74% | 15.46% |
The decision trades how often you rebuild against how long each rebuild takes. Eight drives rebuild twice as often as four and each rebuild is a third as long, so exposure time is roughly a wash and the eight-drive array wins on parity. Fewer bigger drives win on bays, noise, power, spin-up surge and the cost of chassis and cabling.
RAID levels and filesystems: what each one survives and gives up
With 12 TB drives:
| Layout | 4 drives | 6 drives | 8 drives | Survives |
|---|---|---|---|---|
| RAID5 / raidz1 | 36 TB | 60 TB | 84 TB | any 1 |
| RAID6 / raidz2 | 24 TB | 48 TB | 72 TB | any 2 |
| raidz3 | - | 36 TB | 60 TB | any 3 |
| RAID10 / mirrors | 24 TB | 36 TB | 48 TB | 1 per pair |
At four drives, RAID6 and RAID10 cost exactly the same capacity. RAID10 rebuilds faster but survives only the right second failure, where RAID6 survives any. Raidz suits large sequential files; mirrors, whose read IOPS scale with their members, suit virtual machines and databases. RAID5 at modern sizes is defensible only with small drives, a small array and a tested backup. Sizing the array to your data is the storage sizing guide’s subject.
RAID is not a backup. It protects against a drive dying, not against deletion, ransomware, filesystem corruption, a controller writing garbage to every member, a power event, fire or theft. A backup is three copies, on two media, one of them elsewhere; an array is one copy however many drives are in it.
ZFS: the decisions that are permanent
ZFS is the default recommendation for a NAS because of checksums: it can tell you a block is wrong, which no traditional RAID layer can. Whether the memory under it should be ECC is answered in the memory site’s ECC guide, and how much of it each NAS system needs in how much RAM for a NAS or homelab.
ashiftis permanent. Set it too low - typically by letting autodetect believe a 512e drive’s emulated 512-byte sector - and every partial-sector write becomes a read-modify-write for the life of the vdev. Always setashift=12explicitly on a hard-drive pool.recordsizeaffects new files only. It defaults to 128 KiB; for a media NAS,recordsize=1Mon the bulk datasets is the documented recommendation, worth setting before the first file lands.- Width. OpenZFS recommends 3 to 9 disks per raidz vdev, and TrueNAS: “We do not recommend using more than 12 disks per vdev.” Two 6-wide raidz2 vdevs have roughly twice the write IOPS of one 12-wide raidz2 vdev.
- raidz expansion exists since OpenZFS 2.3, but fault tolerance does not change, and blocks written before the expansion keep their original data-to-parity ratio until they are rewritten.
- Fill level. TrueNAS: “At 90% capacity, ZFS switches from performance- to space-based optimization, which has massive performance implications.” Add capacity before 80 percent.
- No hardware RAID underneath: it “will limit opportunities for ZFS to perform self healing on checksum failures.” Use an HBA; the interfaces guide covers IT mode.
mdadm, btrfs, SHR and Unraid
A non-zero mismatch_cnt after an md check on RAID5/6 means data and parity
disagree, and it does not tell you which one is wrong: repair picks parity and
recomputes, which is a guess. That is the strongest argument for ZFS or btrfs over
mdadm, and it is about correctness rather than speed. The
btrfs upstream status page
classifies RAID56 as “unstable” while RAID1, RAID10 and scrub are “OK”. btrfs
RAID1 and RAID10 on a NAS are fine. btrfs RAID5/6 is not a thing to put your array
on.
Synology Hybrid RAID slices each drive at the size boundaries of the drive set and runs a separate RAID group in each layer. With 2 TB + 3 TB + 4 TB + 4 TB it gives 9 TB usable against plain RAID5’s 6 TB: SHR recovers 50 percent more usable capacity from a mismatched set, and with identical drives it is just RAID5. Unraid does not stripe at all: each data disk keeps its own filesystem and a parity disk holds the XOR, so only the disk holding a file must spin up, a file survives whole on its own disk if you lose more disks than parity covers, and the parity disk must be at least as large as the largest data disk. The cost is write throughput, so it suits a media library written rarely and read one file at a time.
SSD caching helps less than it promises
RAM first, L2ARC only when RAM is maxed out, SLOG only for synchronous writes such
as NFS, iSCSI and database commits, so if your NAS serves SMB to
people, an SLOG will do nothing measurable. L2ARC costs RAM for its headers and
wears the SSD. In every OpenZFS release from
2.2.7 (December 2024) to
2.4.4, the latest (August 21, 2026), the zfs.4 manual page lets each 1-second feed
write l2arc_write_max, 32 MiB, plus l2arc_write_boost, another 32 MiB, while the
cache is cold; through 2.2.6 and the 2.1 series both were 8 MiB, and Ubuntu 24.04
LTS still ships 2.2.2. Fed continuously at 32 MiB/s, a consumer SSD rated
for 500 TiB of writes reaches its endurance in roughly six months. Real wear is
usually lower, since a feed writes only what the ARC evicts, though feeds can come
faster than once a second. Losing a special vdev destroys the pool. A media
NAS streaming large files gets approximately nothing from any of them; NVMe caching works through when each one pays.
Choosing between a cache SSD and more RAM,
add RAM.
SMR, shucked drives and mixed batches
SMR does not belong in an array
In 2020 it emerged that drive-managed SMR had been shipped without disclosure across more than one vendor’s consumer lines, and inside a NAS-branded one: the WD Red. Western Digital’s own statement names “Our current device-managed shingled magnetic recording (DMSMR) (2TB, 3TB, 4TB, and 6TB) WD Red series”, and renamed its CMR NAS drives: “WD Red Plus is the new name for conventional magnetic recording (CMR)-based NAS drives”. TrueNAS’s notice gives the tell: the 2, 3, 4 and 6 TB WD Red DM-SMR drives carry the letters EFAX in their product code, the previous-generation EFRX drives in the same capacities (such as WD40EFRX) are CMR, and “The 8TB and larger drives with the ‘EFAX’ code use CMR technology.” That tell covers only the 2020-era WD Red; for other lines a model number does not show the recording technology. TrueNAS also found a ZFS compatibility issue on the WD40EFAX with firmware 82.00A82 under heavy writes, including resilvering.
A drive-managed SMR drive absorbs writes into a media cache and reorganizes them into shingled bands while idle, and a rebuild gives it hours of sustained writes and no idle time. Throughput falls to single-digit megabytes per second and commands start taking seconds. The array does not see a slow drive, it sees a drive that stopped answering, and drops it. Adding a replacement can cost you a second member.
Nor can you test for it easily. /sys/block/sdX/queue/zoned flags only
host-managed drives (and, before Linux 6.8, host-aware ones); drive-managed SMR
reports none, identically to a CMR drive, and no SMART attribute records it. So
filter to CMR in the hard drive listings, and treat an unlabeled listing as
unknown rather than as CMR. Sellers transcribe part numbers wrongly, so confirm an
EFAX or EFRX code on the drive label in the photos. The CMR/SMR
guide covers how shingling works and which current lines use it.
Shucked drives
Shucking an external drive is often the cheapest route to NAS capacity, and it carries two problems that are not the drive’s fault. SATA 3.3 gave pin 3 of the power connector a Power Disable function, so a power supply that delivers 3.3 V on that pin holds such a drive in permanent reset; Western Digital’s own technical brief says plainly that “The HDD is not defective.” Kapton tape over the first three pins or a Molex-to-SATA power adapter fixes it, and the interfaces guide has the details. The second problem is recording technology: an enclosure’s marketing does not state it, and the 2020 episode shows one product line can contain either, so a shucked drive’s CMR status is unknown until it is out of the case, and often not then.
Mixed batches and mixed capacities
Array members share model, firmware revision, manufacturing batch, power-on hours, temperature and workload, and every probability in this guide assumes independence. So buy deliberately mixed, with date codes checked on arrival. A firmware defect is the strongest case for it, being the one failure mode that can take a whole array inside an hour and that no amount of parity survives. Nobody has published a study quantifying how much batch diversity actually buys; it is a cheap hedge against an unquantified risk.
Mixing capacities is a separate question with a per-platform answer:
| Platform | Mixed capacities |
|---|---|
| ZFS raidz | Every drive is used to the size of the smallest. The excess is wasted. |
mdadm |
Same - the array is built on the smallest member’s size. |
| Synology SHR | Layered groups recover most of the excess, as above. |
| Unraid | Fully supported; parity drive must be the largest. |
On ZFS, adding one larger drive to a raidz vdev buys nothing until every member is replaced. That is the single most common mistake in home ZFS builds, and it is why the expansion plan matters more than the drive you buy today.
Helium, power, noise and heat
Helium fill is a per-model manufacturing choice, not a tier, and it does not track capacity cleanly:
| Family | Helium | Air |
|---|---|---|
| WD Red Plus | WD120EFBX (12 TB) only | Everything else, 2-12 TB |
| WD Red Pro | 14 TB and above, plus WD121KFBX (12 TB) | WD122KFBX (12 TB), and 10 TB and below |
| Seagate IronWolf Pro | 12-20 TB, and ST10000NE0008 | 8 TB and below, and ST10000NE0004 |
Every row of that table has a capacity on both sides of the line. The two 10 TB IronWolf Pro part numbers, ST10000NE0004 and ST10000NE0008, are one digit apart and one fill gas apart; so are the 12 TB WD Red Plus pair, WD120EFGX air and WD120EFBX helium, and the 12 TB WD Red Pro pair, WD122KFBX air and WD121KFBX helium.
Helium-sealed drives can expose SMART attribute 22, which starts at 100 and counts down to a threshold of 25; HGST declined to publish the trip points, so any movement below 100 is a signal, and you cannot tell how much margin is left. Not every vendor reports it: Backblaze saw it only on HGST drives, not Seagate’s, so its absence on a helium drive proves nothing. Backblaze’s normalized comparison put helium drives at 1.06 percent AFR against 1.61 percent for comparable air drives, though the helium drives were newer.
One drive reports a different spindle speed from the one it runs at. WD Red Plus footnote 7, for the WD120EFGX:
Actual spindle motor rotational speed for this model is 7200 RPM; although ID Device may report 5400 to reflect previous Performance Class designation.
The drive reports 5400 rpm and physically spins at 7200, so tools reading RPM from it show 5400, and filtering by RPM follows what the listing states; the datasheet footnote settles the spindle.
Idle power sets the bill
A NAS spends nearly all its life idle, so the idle figure is the one that multiplies by 8,760 hours:
| Drive | Idle | Read/write | Standby |
|---|---|---|---|
| WD Red Pro 26 TB | 3.6 W | 6.0 W | 0.3-1.6 W |
| WD Red Plus 2 TB | 2.4 W | 4.0 W | 0.3-1.6 W |
| WD Red Plus 12 TB (WD120EFBX, helium) | 2.9 W | 6.3 W | 0.3-1.6 W |
| WD Red Plus 12 TB (WD120EFGX, air) | 6.1 W | 8.8 W | 0.3-1.6 W |
| Seagate IronWolf Pro 20 TB | 5.5 W | 7.7 W | 0.3-1.6 W |
Look at the two 12 TB WD Red Plus rows. Same family, same capacity, 2.9 W against 6.1 W idle. Over six drives running continuously:
6 x 2.9 W = 17.4 W -> 152 kWh/year
6 x 6.1 W = 36.6 W -> 321 kWh/year
difference 169 kWh/year, from a part-number suffix
The 26 TB Red Pro at 3.6 W idle draws less than the 12 TB air-filled Red Plus at 6.1 W. Bigger drives do not necessarily cost more to run.
Spin-up sizes the 12 V rail, not running load
Datasheet peak 12 V current is 1.2 to 2.1 A per drive:
at 2.0 A per drive on the 12 V rail:
4 drives -> 96 W surge
6 drives -> 144 W surge
8 drives -> 192 W surge
12 drives -> 288 W surge
against idle, at 5.5 W per drive:
8 drives -> 44 W
Eight drives spinning up simultaneously draw roughly four times their idle power, all on the 12 V rail in the first few seconds. That is why NAS chassis and HBAs stagger spin-up, and why a power supply sized on running load browns out at boot: if your array fails to come up after a power cut but works when you start drives by hand, this is why.
Noise and temperature
Uncorrelated sources add at +3 dB per doubling:
| Per-drive seek figure | 4 drives | 8 drives | 12 drives |
|---|---|---|---|
| 28 dBA | 34.0 dBA | 37.0 dBA | 38.8 dBA |
| 32 dBA | 38.0 dBA | 41.0 dBA | 42.8 dBA |
| 36 dBA | 42.0 dBA | 45.0 dBA | 46.8 dBA |
Eight drives are only about 9 dB louder than one, so a drive 4 dBA quieter buys more than halving the drive count. Inside Red Pro the loud end is the air-filled part numbers: WD122KFBX, WD103KFBX and WD102KFBX are all 34 dBA idle against 20 for the quietest helium members.
Seagate’s IronWolf Pro datasheet gives 65 °C as the maximum for 6 to 24 TB, and 60 °C on the 28 to 32 TB models, but adds: “Seagate does not recommend operating at sustained drive temperatures above 60C.” Google’s FAST 2007 study found that over most of the observed range lower temperatures correlated with higher failure rates, probably confounded by the roles cooler drives had in that fleet. Keep drives below 60 °C and stop there. Chasing 30 °C costs fan speed, and fan speed costs throughput you cannot see.
Burn-in, with the commands and the badblocks ceiling
Day-one failures are invisible in every published statistic. Burn-in is the only way you find them, and the window in which finding them is useful is the return window.
badblocks uses a 32-bit block counter, which caps the testable size per block
size. That is why an 18 TB drive returns “Value too large for defined data type”,
and why many people conclude their drive is broken:
| Block size | Maximum testable |
|---|---|
-b 512 (default) |
2.20 TB (2.00 TiB) |
-b 1024 |
4.40 TB (4.00 TiB) |
-b 4096 |
17.59 TB (16.00 TiB) |
-b 8192 |
35.18 TB (32.00 TiB) |
Use -b 4096 up to 16 TiB and -b 8192 above it.
# 1. short self-test, about 5 minutes - catches the obviously dead
sudo smartctl -t short /dev/sdX
# 2. conveyance test, about 2 minutes - looks for shipping damage specifically
sudo smartctl -t conveyance /dev/sdX
# 3. long self-test - full surface read, hours
sudo smartctl -t long /dev/sdX
# 4. record the attributes BEFORE the destructive test
sudo smartctl -A /dev/sdX > /root/sdX.before
# 5. destructive four-pattern write/read - DESTROYS ALL DATA
sudo badblocks -b 4096 -wsv /dev/sdX # up to 16 TiB
sudo badblocks -b 8192 -wsv /dev/sdX # 16 to 32 TiB
# 6. long self-test again
sudo smartctl -t long /dev/sdX
# 7. compare
sudo smartctl -A /dev/sdX > /root/sdX.after
diff /root/sdX.before /root/sdX.after
Step 7 is the actual test. A single SMART reading tells you the drive’s history; the difference across a full-surface write and verify tells you whether the drive is degrading now. Three raw values must be zero on a drive you are keeping:
5 Reallocated_Sector_Ct sectors already remapped
197 Current_Pending_Sector unreadable, not yet remapped
198 Offline_Uncorrectable unreadable during offline scan
Attributes 197 and 198 are the ones that become a failed rebuild: a pending sector is precisely the URE you did not want to meet while degraded. A non-zero raw value on any of the three, on a drive you have owned a week, is a return.
badblocks -w writes and reads four patterns, so a full run moves eight times
the drive’s capacity. A 20 TB Seagate has been reported at 190 to 195 hours,
about eight days, which implies an average rate near the drive’s outer-track figure,
so treat it as a best case.
Budget eight to ten days per large drive. The traffic also counts against the
workload rating; the middle column is the 220 TB/year point at which WD quotes Red
Pro’s MTBF:
| Drive | Traffic per full run | vs 180 TB/yr | vs 220 TB/yr | vs 550 TB/yr |
|---|---|---|---|---|
| 8 TB | 64 TB | 36% | 29% | 12% |
| 12 TB | 96 TB | 53% | 44% | 17% |
| 20 TB | 160 TB | 89% | 73% | 29% |
| 24 TB | 192 TB | 107% | 87% | 35% |
A single burn-in of a 24 TB drive at the 180 TB/year tier exceeds its entire annual workload allowance. That is not a reason to skip burn-in, but it is a reason to do it once rather than routinely. Checking a used drive’s history before you buy it is covered in buying used drives on eBay.
Capacity planning, and the cost of expanding later
What you see is not what is on the box, for two reasons that compound:
6 x "12 TB" in RAID6
raw 72 x 10^12 bytes
usable 48 x 10^12 bytes (4 data drives)
reported 48e12 / 2^40 = 43.7 TiB (binary units)
at 80% 35.0 TiB of actual room
Decimal terabytes against binary tebibytes is the capacity guide’s subject; it costs about 9 percent before the array takes its share. As for that 80 percent, do not plan to run past it: the last fifth of the pool lands on the slow inner part of the platters, copy-on-write filesystems need free runs even to modify a file, and snapshots and a multi-day resilver need room.
Say you need 40 TB now and 80 TB in three years, in an eight-bay chassis:
| Plan | Now | Later | Bays left |
|---|---|---|---|
| 6 x 12 TB RAID6 | 48 TB | add 2 x 12 TB -> 72 TB | 0 |
| 4 x 20 TB RAID6 | 40 TB | add 2 -> 80 TB, add 2 -> 120 TB | 2 |
| 6 x 20 TB RAID6 | 80 TB | nothing needed | 2 |
Bays are the scarce resource, not terabytes. The first plan spends six bays on cheap capacity today and the last two later, and then leaves exactly one move: replace all eight, one full rebuild at a time:
8 drives x 20 TB at ~100 MB/s = 55.6 h each
= 444.8 h
= 18.5 days of degraded or resilvering operation
Eighteen days with no redundancy margin to spare, on drives three years old. That
is the price of “buy small now and upgrade later”, paid in risk rather than money.
raidz expansion softens it a drive at a time, but neither it nor mdadm’s reshape
removes the bay problem.
Price per terabyte argues the other way, and it moves with the market, so the 3.5-inch listings answer it better than a guide can, and Trends shows which way it has been moving. How much storage you need works through sizing the array in the first place.