RAID 50 with 24 disks — 22 usable

RAID 50

Several RAID 5 groups striped together. Each group spends one disk on parity, so more groups means less capacity and more failures survived.

TB on the label, TiB on the screen

• 1 TB × 2422 TB · 20.01 TiB
• 2 TB × 2444 TB · 40.02 TiB
• 4 TB × 2488 TB · 80.04 TiB
• 6 TB × 24132 TB · 120.05 TiB
• 8 TB × 24176 TB · 160.07 TiB
• 10 TB × 24220 TB · 200.09 TiB
• 12 TB × 24264 TB · 240.11 TiB
• 14 TB × 24308 TB · 280.12 TiB
• 16 TB × 24352 TB · 320.14 TiB
• 18 TB × 24396 TB · 360.16 TiB
• 20 TB × 24440 TB · 400.18 TiB
• 22 TB × 24484 TB · 440.2 TiB
• 24 TB × 24528 TB · 480.21 TiB

One TB on the box is 10¹² bytes, while the operating system counts in units of 2⁴⁰ bytes. The names look alike but the units differ, so the same disk reads about 9% smaller. Nothing went missing — the ruler changed.

How the groups are split

• Groups 2 · Disks per group 1222disks · 2
• Groups 3 · Disks per group 821disks · 3
• Groups 4 · Disks per group 620disks · 4
• Groups 6 · Disks per group 418disks · 6
• Groups 8 · Disks per group 316disks · 8

For RAID 50 and 60, how many groups you use is the real decision. Fewer groups spend less on parity and leave more capacity; more groups leave one (or two) spare disks per group, so more failures are survivable when they land well. The guaranteed figure does not change either way — two failures inside one group end it regardless.

Same disk count, other levels

Same level, other disk counts

The rebuild is the dangerous part

When a single-parity array loses a disk, you put a new one in and every remaining disk is read end to end to rebuild it. That is precisely when the survivors work hardest and longest. The bigger and more numerous the disks, the longer the rebuild — days, sometimes — and a second failure during it ends the array. This is why RAID 6 or 10 is preferred over RAID 5 on large arrays.

How to read it

• Usable = total − parity × groups. Every level is that one line.
• RAID 5 spends one parity disk in a single group, so n−1; RAID 6 spends two, so n−2.
• RAID 50 and 60 run several such groups striped together, so the parity is multiplied by the group count.
• RAID 1 holds one copy on every disk, so one disk’s worth; RAID 10 mirrors in pairs, so half.
• Groups must be equal in size, so the group count can only be a divisor of the disk count.
• TB on the box and TiB on the screen are different units — about 9% apart.

RAID is not a backup

RAID protects against exactly one thing: a disk dying. A file deleted by mistake, a file overwritten, ransomware, a failed controller, fire or theft all propagate to every disk at once — a mirror deletes just as faithfully as it writes. A backup is a separate copy on separate hardware, and RAID is not doing that job.