10 × 22 TB in RAID 10: the numbers
| Usable space | 110 TB (100 TB in Windows) |
|---|---|
| Raw space | 220 TB |
| Used for protection | 110 TB |
| Efficiency | 50% |
| Drives that can fail | 1 to 5 |
| Rebuild at 150 MB/s | 40 h |
| Rebuild at 50 MB/s (busy) | 5.1 days |
| Data read during rebuild | 22 TB |
| Read error risk, desktop drives | 83% |
| Read error risk, NAS drives | 16% |
Rebuild
Replacing a failed drive means writing a full 22 TB drive. At 150 MB/s with no other load that takes about 40 h; on a busy array running at 50 MB/s, about 5.1 days. Only the mirror partner is read, so the other drives are not stressed.
Read errors
Reading 22 TB, the chance of at least one unreadable sector is 83% with desktop drives (1 per 1014 bits) and 16% with NAS drives (1 per 1015). Nothing is left to correct it, so the affected file, or on some controllers the whole rebuild, fails. These are worst-case datasheet rates; real drives usually do better.
Our take
With 10 drives, RAID 6 gives more space (176 TB instead of 110 TB) and survives any two failures. RAID 10 is the better pick for databases and virtual machines that write many small blocks.