RAID 1, 5 and 10 on a Dedicated Server: What Each Protects, and What It Doesn't

RAID combines several disks so the server can keep running when one of them fails. That is the whole promise, and it is a useful one. But RAID is often treated as if it also protects data from mistakes, attacks and corruption. It does not. This guide explains RAID 1, 5 and 10, the real risk during a rebuild, how software RAID with mdadm works on Linux, and where RAID stops helping.

What RAID is for

RAID (Redundant Array of Independent Disks) spreads data across disks using mirroring, striping, parity, or a mix. Its main job on a server is availability: a disk dies, the array keeps serving data, and you replace the disk without restoring from backup. Some levels also improve read or write speed.

Keep that word in mind: availability. RAID keeps the server up through a hardware failure. It does not keep a history of your data.

RAID 1: mirroring

RAID 1 writes the same data to two (or more) disks. With two disks you get the capacity of one.

• Survives: the failure of one disk in a two-disk mirror.

• Performance: reads can be spread across both disks. Writes are about as fast as a single disk.

• Rebuild: a straight copy from the healthy disk to the new one. Simple and predictable.

RAID 1 is the usual choice for a two-disk server: the operating system, a web stack, a small database. It is easy to understand and easy to recover.

RAID 5: striping with parity

RAID 5 needs at least three disks. Data is striped across all of them, along with parity information that lets the array rebuild any one missing disk. Usable capacity is the total minus one disk: three 2 TB disks give about 4 TB.

• Survives: one disk failure. A second failure before the rebuild finishes loses the array.

• Performance: good sequential reads. Small random writes are slower, because each write must also update parity (read old data, read old parity, write new data, write new parity).

• Rebuild: the array must read every remaining disk completely to recalculate the missing one.

RAID 5 gives the most usable space for the redundancy. The trade-off is write performance and rebuild risk, covered below.

RAID 10: mirrors, then stripes

RAID 10 needs at least four disks. Disks are paired into mirrors, and data is striped across the pairs. You get half the total capacity.

• Survives: one disk in each mirror pair. Losing both disks of the same pair loses the array.

• Performance: strong reads and writes, with no parity calculation. This is why it is common for busy databases.

• Rebuild: only the partner disk in the affected pair is read, so rebuilds are faster and put less stress on the rest of the array.

Side by side

Level	Minimum disks	Usable capacity	Failures tolerated	Good fit
RAID 1	2	1 disk	1 (in a 2-disk mirror)	OS, web servers, small databases
RAID 5	3	Total minus 1 disk	1	Read-heavy storage, archives
RAID 10	4	50%	1 per mirror pair	Databases, write-heavy workloads

The rebuild is the risky part

When a disk fails, the array runs in a degraded state. In RAID 1 and RAID 5 there is no redundancy left until the rebuild finishes. Several things make this window riskier than it looks:

• The rebuild is heavy work. RAID 5 reads every sector of every remaining disk. Disks that were fine under normal load can fail under this stress.

• Disks often come from the same batch. Same age, same hours, similar wear. When one fails, the others may not be far behind.

• Read errors on large disks. A sector that cannot be read during a RAID 5 rebuild may leave part of the data unrecoverable, because there is no second copy to fall back on.

• Rebuilds take time. On large hard disks, a rebuild can run for many hours, and performance drops while it runs.

As a rule of thumb: RAID 5 on large spinning disks is a weak choice for important data. RAID 10, or RAID 6 (which survives two failures), is safer for big arrays. On NVMe and SSDs rebuilds are much faster, which shrinks the window, but it does not remove it.

What RAID does not protect you from

This is the most important section. RAID copies every change to every disk immediately, including bad changes. It does not protect against:

• Deleting files or dropping a table. The deletion is mirrored instantly.

• Ransomware or a compromised server. Encrypted files are written to all disks.

• Software bugs and bad deploys that corrupt data.

• File system corruption after a crash or a kernel bug.

• Losing the whole server: a controller failure, a failed rebuild, or a problem that affects every disk at once.

RAID is not backup. A backup is a separate copy, ideally on another machine or location, that you can restore from a point in time. A common rule of thumb is 3-2-1: three copies of your data, on two different types of storage, with one copy off-site. Test restores regularly. A backup you have never restored is a hope, not a plan.

Software RAID with mdadm: the basics

On Linux, mdadm manages software RAID. It is widely used on dedicated servers, especially with NVMe disks, and it does not lock you into a specific controller. The CPU cost is small on modern processors.

Create a RAID 1 array from two partitions:

sudo mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/nvme0n1p2 /dev/nvme1n1p2

For RAID 10 with four disks:

sudo mdadm --create /dev/md0 --level=10 --raid-devices=4 /dev/sdb /dev/sdc /dev/sdd /dev/sde

Check the status:

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

A healthy two-disk mirror looks like this. [UU] means both members are up:

md0 : active raid1 nvme1n1p2[1] nvme0n1p2[0]
      937567232 blocks super 1.2 [2/2] [UU]
      bitmap: 1/7 pages [4KB], 65536KB chunk

A degraded array shows [2/1] [U_]. That is your signal to act.

Save the array definition so it assembles at boot:

# Debian / Ubuntu
sudo mdadm --detail --scan | sudo tee -a /etc/mdadm/mdadm.conf
sudo update-initramfs -u

# RHEL / AlmaLinux / Rocky Linux
sudo mdadm --detail --scan | sudo tee -a /etc/mdadm.conf
sudo dracut -f

Replacing a failed disk

Mark the disk as failed and remove it from the array:

sudo mdadm --manage /dev/md0 --fail /dev/nvme1n1p2
sudo mdadm --manage /dev/md0 --remove /dev/nvme1n1p2

After the disk is physically replaced, copy the partition layout from the healthy disk to the new one, then add it back. Check the device names twice: reversing them overwrites the healthy disk.

# copy the GPT layout FROM nvme0n1 (healthy) TO nvme1n1 (new)
sudo sgdisk /dev/nvme0n1 -R /dev/nvme1n1
# give the new disk its own unique GUIDs
sudo sgdisk -G /dev/nvme1n1

sudo mdadm --manage /dev/md0 --add /dev/nvme1n1p2
watch cat /proc/mdstat

If the server boots from these disks, make sure the new disk is bootable too. How you do that depends on whether the server uses BIOS or UEFI boot.

Monitoring

An array that fails silently is not much better than no array. At minimum:

• Set MAILADDR in mdadm.conf so the mdadm monitor service emails you when an array degrades.

• Watch disk health with smartctl -a /dev/nvme0n1 (from the smartmontools package).

• Run periodic consistency checks. Debian and Ubuntu schedule one monthly; you can start one manually with echo check | sudo tee /sys/block/md0/md/sync_action.

Hardware RAID or software RAID?

Hardware RAID controllers handle the array on a dedicated card, often with a battery- or flash-backed write cache. They can help with write-heavy workloads on hard disks. The downside is dependency on that controller model for recovery. Software RAID with mdadm is portable: the array can be read on any Linux system. For NVMe-based servers, software RAID is the common choice. Neither one replaces backups.

Which level should you pick?

• Two disks: RAID 1.

• Four or more disks, database or write-heavy: RAID 10.

• Three or more disks, mostly reads, capacity matters: RAID 5 can work, preferably on SSD or NVMe and with good backups. For large hard disks, consider RAID 6 or RAID 10 instead.

Whichever you choose, add monitoring and a separate, tested backup.

Run it on IM HOST

IM HOST AMD Dedicated and Intel Dedicated servers give you bare metal with ECC memory, NVMe or SSD storage, 10 Gbps networking, DDoS protection and a dedicated IPv4. Disk layouts vary by server, so ask our support team which disk and RAID configurations are available for the model and location you want; we confirm the location before activation. There are no setup fees, our engineers are available 24/7, and our free migration service can move your files, databases and email without downtime.