RAID and the server disk subsystem: levels, fault tolerance, controllers
Servix technical team · Updated
In brief. The RAID level is chosen not by taste but by drive capacity and load type. For databases and 1C — RAID 10, for archives and high-capacity drives — RAID 6, for large arrays — RAID 60. RAID 5 is not built on drives of 4 TB and larger: rebuilding one 16 TB drive takes about 49 hours, and all that time the array runs without redundancy. RAID 4 is not used today — the dedicated parity drive becomes a bottleneck on writes. And most importantly: RAID protects against drive failure and nothing else — a backup is needed separately.
Which level to choose specifically for a storage system — a table by task, plus the distributed RAID of Dell ADAPT and HPE MSA-DP+ — is covered in the article “RAID 5, 6 or 10: what to choose for a storage system.”
Jump to: RAID levels · RAID 4 vs. RAID 5 · how many drives you can lose · why RAID 5 on large drives is dangerous · usable capacity arithmetic · RAID is not a backup · controller: RAID or HBA · cache and power protection · SSD, NVMe, HDD · spare drive · catalog with prices · questions and answers
The disk subsystem is the most common bottleneck of a server and, at the same time, the place where a configuration error leads not to a slowdown but to data loss. We cover RAID levels, the arithmetic of capacity and rebuilds, controller selection and common misconceptions.
RAID levels: what to use and when
Tier | Minimum drives | Fault tolerance | Use case |
|---|---|---|---|
RAID 0 | 2 | no | Only temporary data whose loss does not matter: render cache, intermediate files |
RAID 1 | 2 | one drive of the pair | System volume, hypervisor boot, small servers |
RAID 4 | 3 | one drive | Practically not used: all parity is on one drive, which is also the write bottleneck |
RAID 5 | 3 | one drive | Only small arrays of low-capacity drives or SSDs. Should not be used on high-capacity HDDs |
RAID 6 | 4 | any two drives | Archives, file storage, backups, any arrays of high-capacity drives |
RAID 10 | 4 | one drive in each pair | Databases, 1C, virtualization — anything sensitive to write speed. The best choice for performance |
RAID 50 | 6 | one drive in each group | Large arrays where RAID 5 is split into groups for faster rebuilds |
RAID 60 | 8 | two drives in each group | Large storage with dozens of drives: a compromise between capacity and risk |
A simple rule for choosing: if the load is small-block writes (database, 1C, mail server), take RAID 10 and do not skimp on drives. If the load is sequential reads and storage of large files, take RAID 6 and save on the number of drives.
RAID 4 vs. RAID 5: why the fourth level is missing from configurators
There is exactly one difference between them — where the parity is stored. In RAID 4, a separate drive is dedicated to it; in RAID 5, parity blocks are distributed across all drives of the array.
The consequence is fundamental. Any write to a RAID changes both the data and the parity. In RAID 4, every write operation, whichever drive it goes to, must update the same parity drive — and it instantly becomes a bottleneck: the other drives sit idle, and the array runs at the speed of a single drive. In RAID 5, parity is spread across all drives, the load is distributed, and the array is faster on writes with the same redundancy and the same usable capacity.
That is why server controllers do not offer RAID 4: with the same capacity loss, it loses on speed. The only place where the dedicated-parity scheme is still alive is specialized file systems such as NetApp's WAFL, where the drawback is offset by write journaling and non-volatile memory. In an ordinary server with a Smart Array or PERC controller, there is no choice between 4 and 5: you take 5, and on high-capacity drives, 6.
Fault tolerance in numbers: how many drives you can lose
The phrase “survives a drive failure” without specifying which drive is misleading. The correct wording is:
- RAID 5 — any one drive. A second failure before the rebuild finishes means complete loss of the array.
- RAID 6 — any two drives. This is a fundamentally different level of protection: the array survives the failure of a second drive exactly when it is most likely — during the rebuild of the first.
- RAID 10 — one drive in each mirrored pair. In an array of eight drives, up to four can fail — if they are from different pairs. But two drives of the same pair kill the entire array, so “withstands four failures” is a lottery, not a guarantee.
- RAID 50 — one drive in each RAID 5 group. An array of two groups of four drives will survive two failures in different groups and will not survive two failures in the same group.
- RAID 60 — two drives in each group. The most resilient scheme for arrays of dozens of high-capacity drives.
The practical point of RAID 50 and 60 is not fault tolerance as such but rebuild speed: when a drive fails, only its group is recalculated, not the entire array. On twenty drives, this is the difference between a day and a week in a degraded state.
Why RAID 5 on large drives is a bad idea
When a drive fails, the array starts rebuilding onto a spare, and this is not an instant operation. A 16 TB drive at a real-world rebuild speed of about 90 MB/s under working load is recalculated in roughly 49 hours — almost two days, during which the array runs without any redundancy.
There is a worse problem. To rebuild one drive, the controller must read all the others in full: in an array of 8 drives of 16 TB each, that is 112 TB of continuous reading. For enterprise SAS drives, the rated probability of an unrecoverable read error is one per 1015 bits, that is, roughly once every 125 TB read. For consumer-class SATA drives it is an order of magnitude worse, one per 1014 bits, about 12.5 TB. Compare this with the 112 TB that has to be read, and it becomes clear that for SATA, a RAID 5 rebuild at such volumes will almost certainly hit a read error, which means either loss of the array or loss of part of the data.
Add to this that the drives in a server are usually from the same batch and have the same operating hours: if one has failed from wear, the rest have reached the same threshold. That is why RAID 6 is used for high-capacity drives — a second parity drive costs less than restoring data from a backup, if one exists at all.
Usable capacity arithmetic
Let's calculate for 8 drives WD Ultrastar DC HC550 16 TB SAS at 5,031,764 UZS each — 40,254,112 UZS for the set:
- RAID 6 — 96 TB usable. Two drives go to parity; it withstands two failures. A workable option for file storage and backups.
- RAID 10 — 64 TB usable. Half of the capacity goes to mirrors, but writes are many times faster and a rebuild takes hours, not days: one drive is copied rather than the entire array being recalculated.
- RAID 5 — formally 112 TB, that is, 16 TB more than RAID 6. Exactly those 16 TB are the price of the risk described above. On such drives, it is not used.
And a separate note on “usable”: this is capacity before formatting. The file system, snapshots and service areas will take a few more percent, and the array must not be filled to the brim: above 80–85% utilization, speed drops and rebuilds become harder. Plan with headroom.
RAID is not a backup
A misconception that regularly leads to losses. RAID protects against physical drive failure and nothing else. It will not save you from:
- file deletion by a user or an administrator error;
- ransomware — the data will be neatly encrypted on all drives of the array at the same time;
- file system or database corruption;
- controller failure, fire, server theft.
The backup must be stored outside this server, and at least one copy of it must be out of reach of the production network so that it is not encrypted along with the original.
Controller: RAID or HBA
- RAID mode — the controller builds the array and presents a single volume to the operating system. The classic scenario for a standalone server. Examples: Dell PERC H730 (5,750,588 UZS), HPE Smart Array MR416i-a (8,625,882 UZS).
- HBA mode (pass-through) — the controller presents the drives to the system directly. It is needed for software-defined storage and hyperconverged clusters that manage redundancy themselves: Ceph, vSAN, Storage Spaces Direct, ZFS. Hardware RAID under such solutions is not just unnecessary but harmful — it hides the drive status from the system. Suitable options are HPE Smart Array E208i-a (5,750,588 UZS) or Dell PERC H345 (4,312,941 UZS).
Decide on the mode before ordering: not all controllers work equally well in both. Model numbers are decoded in the reference guide to PERC and Smart Array controllers.
A separate topic is the number of drives. A controller with eight lanes addresses eight drives directly; to connect twenty-four, you need a backplane with an expander or a separate expander card (7,188,235 UZS).
Cache and power protection
Controllers with write cache provide a significant performance boost but require a battery or a supercapacitor. Without cache power protection, a power outage means losing the data that was in the cache at the moment of failure — with a high probability of database corruption.
This is also the cause of the most common “unexplained” degradation: after three to four years the battery loses capacity, the controller switches the cache to write-through mode on its own, and the server starts running several times slower. There is not a single error in the log — only an entry about the status of the cache power module. If “the server suddenly became slow” and nothing was changed, check the battery first.
SSD, NVMe, HDD — what goes where
- NVMe — databases, heavily loaded applications, the storage cache tier. Example: Micron 7400 Pro 1.92 TB U.3 — 4,312,941 UZS.
- SAS/SATA SSD — virtualization, general-purpose tasks, system volumes. Example: Intel D3-S4520 7.68 TB — 10,782,352 UZS.
- HDD — archives, backups, file storage with large capacity and low activity. Here the cost per terabyte still decides: WD Ultrastar 22 TB SAS costs 8,625,882 UZS, which is noticeably cheaper per terabyte than any SSD.
Drives of different types, capacities and spindle speeds must not be mixed in one array: the controller will level the entire array down to the worst drive. For the system and hypervisor, it is better to allocate a separate pair — for example, the boot module HPE NS204i-p (10,063,529 UZS) or Dell BOSS (7,907,058 UZS) — so as not to occupy drive bays.
Spare drive
The replacement drive must be on site before the failure, not ordered after it. Options:
- Hot spare. The drive sits in the chassis, spun up and waiting. When a failure occurs, the rebuild starts automatically, without human involvement — at a remote site, this is the only sensible option.
- Global spare. One spare drive serves several arrays on the same controller. It saves a bay, but if there are many arrays, the risk of two failures competing for one spare is real.
- Cold spare. The drive sits on a shelf. It is cheaper, but the rebuild will start only after someone arrives and inserts it. For a server in another city, that adds a day to the time without redundancy.
The spare drive should be the same model and the same capacity as the array drives. A larger drive will work, but the extra capacity will be lost; the controller simply will not accept a smaller drive.
Controllers and drives in the catalog
Item | Use case | Price |
|---|---|---|
RAID 5/6/10, 4 GB of protected cache | 8,625,882 UZS | |
RAID with cache and battery, PowerEdge 15G | 8,625,882 UZS | |
RAID with cache, PowerEdge 13–14G | 5,750,588 UZS | |
no cache, HBA mode | 5,750,588 UZS | |
no cache, simple volumes and pass-through | 4,312,941 UZS | |
connecting a 24-drive chassis | 7,188,235 UZS | |
2 × 480 GB NVMe boot mirror | 10,063,529 UZS | |
databases, cache tier | 4,312,941 UZS | |
virtualization, general-purpose volume | 10,782,352 UZS | |
RAID 6 array, file storage | 5,031,764 UZS | |
archive, backups | 8,625,882 UZS |
What to check before building an array
- Load type: small-block writes — RAID 10; sequential reads and storage — RAID 6.
- Capacity of a single drive: from 4 TB and up, RAID 5 is not considered.
- Whether there is a spare drive on site and what kind — hot or cold.
- The condition of the controller battery or supercapacitor, especially if the server is more than three years old.
- Controller mode: RAID or HBA — software-defined storage needs the latter.
- A separate volume for the system so that the hypervisor does not occupy array bays.
- Where and how often the backup goes — and whether restoring from it has been tested.
- Array utilization: plan for no more than 80% of usable capacity.
We will build a disk subsystem for your task: calculate usable capacity, select the RAID level, a controller with the right cache, and drives with the right headroom. Write to us; drives and controllers with prices are here.
Questions and answers
What is the difference between RAID 4 and RAID 5?
The location of the parity. In RAID 4 it lies entirely on one dedicated drive; in RAID 5 it is distributed across all drives of the array. Their usable capacity and fault tolerance are the same — one drive — but in RAID 4 the parity drive is updated on every write and becomes a bottleneck, so in practice RAID 5 is used.
What is RAID 5 in simple terms?
An array of three or more drives in which the data is distributed across all drives along with checksums that make it possible to restore the contents of any one failed drive. Usable capacity equals the total minus one drive. The price of the savings is a long rebuild and complete loss of the array if a second drive fails before the rebuild finishes.
How fault-tolerant is RAID 50?
RAID 50 is striping across several RAID 5 groups. One drive can fail in each group; the array will survive two failures in different groups, but not two in the same group. The main gain is not fault tolerance but the fact that, on failure, only one group is recalculated, and the time in a degraded state is reduced several-fold compared with a wide RAID 5.
How many drives are needed for RAID 6?
At least four; a sensible minimum is six. With four drives, half of the capacity goes to parity, and in terms of the capacity-to-speed ratio RAID 10 is more advantageous. RAID 6 starts to make sense with six or more drives, where the share lost to parity decreases.
Which stripe size should I choose?
The controller's default value (usually 64–256 KB) suits most tasks. A smaller stripe benefits databases with small random operations; a larger one benefits streaming writes of video and archives. Changing it makes sense only when there is a measured problem: after the array is created, the stripe size cannot be changed without rebuilding it.
Can drives of different capacities be put into one array?
Technically yes, but the controller uses only the capacity of the smallest drive from each drive — the difference will be lost. Drives with different spindle speeds or of different types (SSD and HDD) should not be mixed in one array: performance will be leveled down to the worst.
Related reading
RAID 5, 6 or 10: what to choose for a storage system and why it depends on the drives
Why RAID 5 is dangerous on high-capacity drives, why RAID 10 is faster on writes, and how to choose the array level for HDD, SSD and NVMe. Practical recommendations with calculations.
Server drives: HDD or SSD, SAS, SATA or NVMe — what to buy for the DL380 and R750
Which drives to buy for an HPE DL380 or Dell R750 server: SAS, SATA, NVMe, PERC and Smart Array RAID controllers. Prices in UZS, three ready-made sets. Tashkent.
SFF or LFF: which drive bays to choose in a server
What 8SFF, 12LFF and 24SFF mean, which drives fit in 2.5″ and 3.5″ bays, when to choose LFF, when SFF and what NVMe requires. DL380 and R760 with prices.
Database server for PostgreSQL, MS SQL and 1C: how to choose and how much it costs
Server for a DBMS: MS SQL per-core licenses, memory for the database size, SSDs in RAID 10, the log on a separate volume, battery-backed cache. Builds and prices as of November 2026.



