RAID Levels Explained: RAID 0, 1, 5, 6 and 10 for Servers

Applies to: dedicated servers with two or more disks; Linux software RAID (mdadm 4.4) and hardware RAID

RAID combines several disks into one volume, either to survive a disk failure, to be faster, or both. The RAID level decides which. RAID is not a backup: it doesn’t protect against deleted files, ransomware or a server fire, so keep backups as well.

The common levels

• RAID 0 (striping): data is spread over all disks. Fastest and uses all the space, but one failed disk loses everything. Only for temporary data. Minimum 2 disks.
• RAID 1 (mirroring): every disk holds the same data. Survives the loss of all but one disk; usable space is one disk. The usual choice for servers with 2 disks.
• RAID 5 (striping with parity): survives one failed disk; usable space is all disks minus one. Minimum 3 disks. Rebuilds on large disks are slow and stressful, so it is less popular today.
• RAID 6 (double parity): survives two failed disks; usable space is all disks minus two. Minimum 4 disks.
• RAID 10 (mirrors, striped): pairs of mirrors combined for speed. Survives one failed disk per mirror; usable space is half. Minimum 4 disks. A good choice for databases.

Which one for a server?

• 2 disks: RAID 1.
• 4 or more disks, speed matters (databases, virtual machines): RAID 10.
• Many large disks, capacity matters (backups, archives): RAID 6.

Software or hardware RAID?

Software RAID is built into Linux (mdadm) and costs nothing; it is common on dedicated servers. Hardware RAID uses a controller card with its own tools. Ask for the setup you need when you order or reinstall a dedicated server.

Check a Linux software RAID

cat /proc/mdstat
sudo mdadm --detail /dev/md0

In /proc/mdstat, [2/2] [UU] means both disks of a mirror are working; [U_] means one is missing. We tested this on a Linux server with a practice RAID 1 array: after marking one disk as failed, mdadm --detail showed State : clean, degraded; after adding a replacement disk it rebuilt and returned to clean with [UU].

sudo mdadm /dev/md0 --fail /dev/sdb1 --remove /dev/sdb1    # take out a failing disk
sudo mdadm /dev/md0 --add /dev/sdc1                        # add the replacement (it rebuilds)

On a Ucartz dedicated server, open a ticket if a disk fails: we replace the hardware, and you (or our engineers) add the new disk back to the array. First steps with a new server: dedicated server checklist.