When you order a dedicated server with two or more disks, someone asks which RAID level you want. Picking the right one takes two minutes once you know what each RAID level does to space, speed and safety.
Here are the RAID levels you will be offered on a dedicated server, in plain terms: what each one does to space, speed and safety, and which to choose.
First: RAID is not a backup
RAID keeps the server running when a disk dies. It does nothing when you delete a folder, when ransomware encrypts the data, or when a bad deploy corrupts the database, because it faithfully copies the mistake to every disk at once. Keep backups somewhere else as well. Ours is an add-on with the last 48 hours retained; keep your own off-site copy too.
The RAID levels compared
| Level | Minimum disks | Usable space | Failures survived | Good for |
|---|---|---|---|---|
| RAID 0 | 2 | All of it | None | Scratch space, caches you can lose |
| RAID 1 | 2 | Half | 1 | Almost every two-disk server |
| RAID 5 | 3 | All but one disk | 1 | Read-heavy storage on smaller disks |
| RAID 6 | 4 | All but two disks | 2 | Large storage arrays |
| RAID 10 | 4 | Half | 1 per mirrored pair | Databases, virtualisation, busy servers |
RAID 0 stripes data across disks for speed. Lose one disk and everything is gone. Never use it for anything you need.
RAID 1 writes the same data to two disks. One can die and the server carries on; replace it and the mirror rebuilds. Reads can be a little faster, writes are the same as one disk.
RAID 5 stripes data with parity across three or more disks. Any one disk can fail. Writes are slower because parity is recalculated, and rebuilding a large disk takes long enough that a second failure during the rebuild is a real risk.
RAID 6 is RAID 5 with two parity blocks, so any two disks can fail. Slower writes again, safer rebuilds. The right choice for big arrays of large disks.
RAID 10 mirrors pairs of disks and stripes across the pairs. Fast reads and writes, survives one failure per pair, and rebuilds quickly because it only copies a mirror. The default for databases and anything write-heavy.
Hardware or software RAID
A hardware RAID controller does the work on a card, often with a battery-backed cache. Linux software RAID (mdadm) does it in the kernel and, on modern CPUs, is as fast for most workloads and easier to move between machines. Either is fine on a rented server; what matters is that someone is watching for a failed disk.
Which to choose on a rented dedicated server
Two disks: RAID 1, almost always. Four disks running a database or virtual machines: RAID 10. Four or more large disks for storage or backups: RAID 6. Avoid RAID 5 on large SSDs for anything important; the rebuild window is the risk. Our Arya and Shakti builds ship with two SSDs, which we set up as RAID 1 unless you ask for something else.
Watch the array
A RAID that degrades silently is a single disk waiting to fail. Set up alerts: mdadm --monitor with an email address for software RAID, or the controller's tool for hardware RAID, and check the hardware event log in IPMI now and then.
Questions we get asked
Can I change RAID level later? Usually only by rebuilding the array, which means backing up and reinstalling. Choose at order time.
Who replaces a failed disk? On a rented server, we do, at no charge. You tell us, or we see it in monitoring.
Is RAID 1 slower than a single disk? Writes are about the same, reads can be faster.
Should I use RAID with NVMe? Yes, the same rules apply; mirroring protects against a failed drive regardless of type.
Do VPS plans use RAID? Yes, our VPS nodes run on NVMe in RAID; that is our job, not yours.
