RAID on a dedicated server: 1, 10 or none
Almost every dedicated server comes with two or four disks, and the host asks how to set them up. This decides whether the server survives a disk failure, how much space is left for data and how fast the database will run. Below: which RAID level suits which task, how software RAID differs from hardware RAID and what to check right after the server is delivered.
The short answer. On two disks, use RAID 1, a mirror: the server keeps running if one disk fails. On four or more disks for a database or virtual machines, use RAID 10: the same protection and almost twice the write speed. Go without RAID only if the data does not live on this server: it is replicated to other machines or you can afford to lose it. And in every case you need backups: RAID protects against disk failure, not against deleted files.
RAID levels in one table
| Level | Disks | Usable capacity | Survives the failure of | When it fits |
|---|---|---|---|---|
| No RAID | 1 or more | All disks | Nothing | Nodes whose data is held on other servers |
| RAID 0 | 2 or more | All disks | Nothing, the whole array is lost | Temporary data where only speed matters |
| RAID 1 | 2 | One disk | One disk | Almost any server with two disks |
| RAID 10 | 4 or more, an even number | Half | One disk in each pair | Databases, virtualization |
| RAID 5 | 3 or more | All but one | One disk | Archives on SSD with few writes |
| RAID 6 | 4 or more | All but two | Two disks | Large storage on HDD |
An example: four 1 TB disks give 2 TB for data in RAID 10, 3 TB in RAID 5 and 2 TB in RAID 6.
RAID 1: the default choice
Two disks hold the same data. If one fails, the server keeps running on the other, and once the disk is replaced the array rebuilds itself. Reads can come from both disks at once, while writes run at the speed of one. This is how the most popular budget plans are built. A few examples as recorded in the accsly catalog on 14 November 2026:
- Hetzner AX41: 2×512 GB NVMe, €59 a month excluding VAT;
- OVHcloud Rise RISE-1: 2×512 GB NVMe, software RAID, €56.99 a month excluding VAT;
- Beget Blade №4245p-64: 2×960 GB NVMe in RAID 1, ₽31,400 a month.
In the catalog, the disk capacity of such plans is the total: AX41 is recorded with 1,024 GB, while RAID 1 leaves 512 GB for data. Keep this in mind when comparing plans by capacity. How to read the rest of the configuration line is covered in the article how to read a dedicated server configuration.
RAID 10: when you have four disks
RAID 10 is two or more mirrored pairs with the data striped across them. Writes go to all pairs in parallel, so databases and virtual machines run noticeably faster than on RAID 1. The array survives the failure of one disk in each pair, but not of both disks in the same pair. Examples from the catalog:
- Cloverhost: Intel Xeon E3-1230, 16 GB, 4×1 TB SATA in RAID 10, $69 a month;
- Digital Pacific X-C6620-2: 4×480 GB SSD in RAID 10, the host states plainly "capacity after RAID 960 GB", AUD 473 a month;
- HostingSource: hardware RAID 10 on SSD across the whole line-up, for example a Xeon E5-2680 v4 with 4 TB for $139 a month.
RAID 5 and 6: why they are chosen less often
RAID 5 and 6 give more space, but every write requires recalculating parity, so writes are slower than on RAID 10. The main risk is the rebuild. After a disk fails, the array reads all the remaining disks in full; on large HDDs that takes hours or days, and throughout that time a second failure (for RAID 5) means data loss. The Linux kernel documentation specifically warns that if a RAID 5 or RAID 6 array is both degraded and was shut down uncleanly, data may be silently corrupted. That is why large HDD storage uses RAID 6, and databases use RAID 10.
When you can go without RAID
- The server is one of several identical machines and the data is stored on the others: a database replica, a cluster node, a build server.
- The storage itself keeps copies on different servers, as distributed storage systems do.
- The server holds only temporary data: a cache, working files for calculations. Here RAID 0 is sometimes used for speed, with the understanding that the failure of any disk destroys the whole array.
In all other cases, a disk failure without RAID means reinstalling the server and restoring from a backup, which is hours of downtime.
Software or hardware RAID
Software RAID in Linux is assembled by the kernel (the md subsystem and the mdadm utility). It does not depend on a controller: the array's disks can be moved to any server running Linux and read there. On modern processors its overhead is barely noticeable. Hardware RAID is handled by a separate controller. Its advantage is a write cache backed by a battery or flash memory, which speeds up writes and preserves them during a power loss. The downside is that the array is tied to the controller model, and the disks cannot always be read on a different one.
In the accsly catalog records, RAID is named for 806 plans at 54 of the 197 hosts: RAID 1 for 193, RAID 10 for 32, RAID 0 for 17, RAID 5 and 6 for 6. Software RAID is stated explicitly for 412 plans and hardware RAID for 51. For the remaining plans the RAID level is not stated, and you need to ask.
What to check once the server is delivered
- Array status. For software RAID:
cat /proc/mdstatandmdadm --detail /dev/md0. All disks should be active and the sync should be complete. - Disk health.
smartctl -afrom the smartmontools package: power-on hours, reallocated sectors, SSD wear. This shows whether you were given new disks and whether any of them is due for replacement. - Alerts. An array that has lost a disk keeps working, and without alerts you may not find out for months.
mdadm --monitorsends an email when a disk fails; you can also hook up general monitoring, see the article server monitoring. - Disk replacement. Ask the host how quickly a disk is replaced, whether the server has to be shut down and whether replacement is paid.
RAID is not a backup
RAID instantly repeats any change on all disks, including a deleted file, a failed update and data encrypted by malware. Nor will it save you if the controller fails, if there is a fire or if your account is blocked. You need separate copies in another place: how to set them up is covered in the article server backups, and how a backup differs from a snapshot in the article snapshots and backups.
Selections of servers with several disks: dedicated servers for data storage and dedicated servers with NVMe.