RAID 5 vs RAID 6: The Rebuild Window Decides It

This comparison is not really about capacity. RAID 6 costs exactly one more drive of parity than RAID 5, and what it buys is survival of the rebuild window. Everything else is arithmetic around that single point. At four bays with 8 TB drives, a single-parity rebuild reads 24 TB and is done inside a day, so RAID 5 is a defensible choice. At eight bays with 20 TB drives the same rebuild reads 140 TB off seven survivors that have no protection left, for 37 to 69 hours. Same layout, different decade of drive capacity, completely different answer.

What is the difference between RAID 5 and RAID 6?

RAID 5 stripes data across every member and dedicates one drive worth of space to parity, so it tolerates exactly one drive failure. RAID 6 dedicates two drives worth and tolerates two. Usable capacity is (n minus 1) times the smallest drive for RAID 5 and (n minus 2) times the smallest drive for RAID 6. Both truncate every member to the size of the smallest drive in the set, so a single small drive in an otherwise large array strands capacity on every larger member.

ZFS calls the same two schemes RAIDZ1 and RAIDZ2, and Synology calls them SHR and SHR-2. The parity counts are identical, and so is the argument on this page. The mechanics of each level are covered in RAID levels explained, and the mixed-drive behaviour that separates SHR from the rest is worked band by band in SHR against RAID 5.

The reason the choice is interesting is that the two levels differ by one drive of capacity, a fixed and easily priced cost, while the protection difference is not fixed at all. It grows with member count and with drive size, and both of those have grown enormously since the rule of thumb most people are carrying around was formed.

How much usable capacity does the second parity drive cost?

Exactly one drive, always. Not a percentage, one drive. That is why the cost of double parity falls sharply as the array gets wider: at four bays it is a quarter of your raw capacity, and at twelve bays it is a twelfth.

Computed with the same functions behind the RAID capacity calculator. Figures are decimal TB as printed on the drive label. Your operating system reports each figure about 9.05 percent smaller because it counts in binary tebibytes, so 120 TB usable reads as 109.1 TiB. That is explained in full on the TB against TiB chart.

Read the efficiency column downward. Double parity on four bays returns half your raw capacity, which genuinely stings. Double parity on twelve bays returns 83 percent, which is barely worse than the 92 percent single parity would have given. The wider the array, the cheaper the insurance and the more you need it, which is a rare case of the incentives pointing the same way.

What actually happens during a rebuild?

A rebuild reconstructs the contents of the failed drive by reading every surviving member end to end and recomputing the missing blocks. There is no shortcut, because parity is only meaningful across the full stripe. The array reads (n minus 1) times the drive size under single parity, and (n minus 2) times under double parity, and writes one drive worth of reconstructed data onto the replacement.

Two properties of that operation are what make it the dangerous part of owning an array. First, under RAID 5 the array has no redundancy at all while it runs. A second failure during the rebuild is total loss. Under RAID 6 a second failure is survivable and the rebuild simply continues at reduced protection. Second, it is the single heaviest workload the drives ever see: every surviving drive reads its entire surface, at maximum rate, for hours or days, at the same time, on drives that were bought in the same order and have run the same hours.

Correlated age is the underappreciated part. Large-fleet annualised failure rates for modern 3.5 inch drives sit around 1.0 to 1.6 percent a year and rise past roughly five years of service. Those are independent-drive figures. Eight drives from one order have the same firmware, the same manufacturing window and the same running hours, so their failures are not independent, and the rebuild is exactly the moment that stops being an academic point. Split a large drive order across two batches or two retailers.

How long does a rebuild take, and how much does it read?

Rebuild duration tracks the size of one member rather than the size of the whole array, because the survivors read their own surfaces in parallel while the replacement is written. Realistic resilver rates on a NAS that is also serving files land between 80 and 150 MB per second.

Rebuild reads is the volume the array must pull off the surviving members to reconstruct one failed drive. The URE columns apply the manufacturer rating to the single-parity read volume, at the two ratings NAS drives are sold against. Read the caveat below before treating either column as a prediction. Precomputed durations for more capacities are on the rebuild time chart.

Two columns carry the argument. The read volume climbs with both member count and drive size, so it compounds: an eight-bay array of 20 TB drives reads 140 TB under single parity against 120 TB under double, and does it with no protection left. And the duration column shows that this is not an hour of nervousness, it is two to three days.

What is a URE and should the 10^14 figure scare me?

An unrecoverable read error is a sector the drive cannot read back correctly even after retries and error correction. Manufacturers publish a rate for it, expressed as fewer than one such sector per 10^14 or 10^15 bits read. NAS and enterprise drives are generally rated at the better of the two.

Treat that number as a warranty floor, not a measurement. It is the worst case the manufacturer will stand behind, not the rate the drives actually exhibit, and observed rates across large fleets run substantially better. Modern software also survives the event: ZFS reports which file was affected and carries on, and mdadm can be instructed to continue rather than dropping the array. The old claim that RAID 5 rebuilds are statistically doomed above a certain drive size takes the floor figure as a forecast, which it is not. Use the URE columns to rank layouts against each other, and never to predict that your particular rebuild will fail.

Ranking is still the useful thing it does. Whatever the true rate is, it applies per bit read, so an array that reads 140 TB to recover is roughly six times more exposed than one that reads 24 TB, and RAID 6 removes the consequence entirely by leaving a second parity syndrome to reconstruct from. That relationship holds no matter which absolute rate you believe. Work your own numbers through the drive failure probability calculator if you want the whole-array view over five years rather than the rebuild view.

When is RAID 5 still the right choice?

At four bays, with drives of roughly 8 to 10 TB or smaller, on an array that is also backed up. That is the honest range, and it is a real range rather than a grudging concession. A four-bay array of 8 TB drives rebuilds by reading 24 TB in about a day, and single parity returns 24 TB usable against 16 TB for double, which is a 33 percent capacity difference on a small array where capacity is tight.

The matrix below applies that rule across the grid, using rebuild read volume as the ranking variable because it is the thing that actually changes.

Ranked on single-parity rebuild read volume, with an additional rule that arrays of eight or more members never earn an unqualified single-parity verdict, because member count multiplies the number of drives that must all read perfectly. Run your own drive set through the RAID capacity calculator for the exact figures behind any cell.

Notice where the boundary falls. A four-bay array does not need RAID 6 until the drives get large, and at 4 TB or 8 TB it genuinely does not. An eight-bay array is on the wrong side of the line almost immediately, which is why the eight-bay buying guide treats double parity as mandatory rather than as an option.

What does the second parity drive cost in money?

One drive at whatever you paid for it, and that is a number worth looking at directly rather than as a percentage, because it is usually smaller than the anxiety around it.

Prices are the verified figures used across this site and move often. Cost per usable TB is the whole drive spend divided by usable capacity, which is the number that lets you compare a six-bay build against an eight-bay one honestly.

The cost per usable terabyte columns are the ones to compare. On a four-bay build the second parity drive raises the cost of every usable terabyte by half, which is a real decision. On an eight-bay build it raises it by a third, and on twelve bays by a fifth. If you are choosing between a wider array with double parity and a narrower one with single parity at the same money, the wider array with RAID 6 usually wins on both capacity and protection, which is the least intuitive result on this page.

Is RAID 6 slower, and does that matter at home?

Sequential reads are effectively identical, sequential writes are slightly slower, and random small writes are meaningfully slower. RAID 6 computes two parity syndromes rather than one, so a small write that does not cover a full stripe has to read and rewrite both, which is roughly a six operation penalty against four for RAID 5.

On a home network almost none of that is visible, because the link is the limit long before the array is. A single 7200 rpm drive sustains 150 to 260 MB per second and already saturates gigabit at 113 MB per second, and even a 2.5GbE link at about 280 MB per second is filled by any four-drive parity array. Whether the array could have delivered 620 or 700 MB per second is irrelevant when the cable caps it at 280. That relationship is worked through in 2.5GbE against 10GbE.

If your workload really is random small writes, virtual machine disks or a database rather than files, neither level is the right answer and the comparison you want is RAID 10 against RAID 6. Adding an NVMe cache drive in front of a parity array helps with reads and with metadata, and it does not change the parity arithmetic at all.

Which should you choose, by reader type?

Five readers, five answers, and none of them is that it depends.

• Four bays, drives of 8 TB or smaller, backed up elsewhere. Winner: RAID 5. The rebuild reads 24 TB and takes about a day, and you keep 24 TB instead of 16 TB. What you give up is the ability to survive a second failure during that day, which is why the backup is part of the condition rather than a footnote.
• Four bays, drives of 12 TB or larger. Winner: RAID 6. The rebuild moves from hours into days, and on a four-bay unit double parity also means you can lose a drive and still not be racing the clock. Buy larger drives rather than more of them if the capacity hurts.
• Six to eight bays, any modern drive size. Winner: RAID 6, and it is not close. The capacity cost has fallen to a sixth or an eighth of raw, and the read volume during a rebuild has climbed past 84 TB. You give up one drive of space and remove the entire class of failure that kills arrays.
• Twelve bays or more. Winner: RAID 6 at minimum, and consider RAIDZ3. Double parity costs 8 percent of raw here. A third parity drive costs another 8 percent and starts to look reasonable on an archive you rarely rewrite.
• You have no backup at all. Winner: neither, fix the backup first. Parity protects against a drive dying. It does nothing about deletion, ransomware, a failed controller writing garbage, theft or fire. A RAID 6 array with no second copy is worse protected than a RAID 5 array that is backed up nightly. Start with the 3-2-1 backup rule.

One practical note that sits underneath all five. If you are still choosing hardware, the bay count you buy decides this question more than the layout you pick later. A four-bay unit makes RAID 5 a live option for years. An eight-bay unit settles the question the day you fill it.

How do I check these numbers for my own array?

The arithmetic is deliberately simple so you can verify it. Take the smallest drive in the set. Usable capacity is that size times (drives minus 1) for single parity, or times (drives minus 2) for double. Rebuild reads is the same multiplier applied to the surviving members. Rebuild duration is one drive size divided by your resilver rate, so a 12 TB member at 100 MB per second is 12,000,000 MB divided by 100, which is 120,000 seconds, or about 33 hours.

Run the exact set through the RAID capacity calculator, which evaluates RAID 5, RAID 6, RAIDZ1, RAIDZ2, RAIDZ3, SHR, SHR-2 and the rest against the same drives at the same time, including mixed sizes. If the figure your NAS reports is about 9 percent below what you compute, that is decimal against binary units and nothing is wrong. If it is 25 percent below, you have a filesystem reserve or a spare assigned, which is a different question worth chasing down.

Related reading

• RAID levels explained for the mechanics of every level in one place
• RAID 10 against RAID 6 when the workload is random writes rather than files
• SHR against RAID 5 for the mixed drive size case
• Rebuild time chart for precomputed durations at every capacity
• How to choose NAS drives because the drives are two thirds of the spend

Deciding on a specific model? We review the Western Digital Red Pro 24TB NAS hard drive review, Seagate Exos 20TB enterprise hard drive review and Seagate IronWolf Pro 16TB review in full.