Introduction
A RAID (Redundant Array of Independent Disks) controller is one of the most critical components in any server build. It determines how your data is stored, protected, and accessed. Choosing the wrong RAID level or controller can lead to poor performance, data loss, or unnecessary cost. This guide breaks down everything you need to know.
RAID Levels Explained
RAID 0 — Striping (No Redundancy)
- How it works: Data is split across multiple drives for maximum speed
- Pros: Highest read/write performance, full capacity utilization
- Cons: Zero fault tolerance — one drive failure loses all data
- Best for: Temporary scratch storage, video editing workstations (not production servers)
RAID 1 — Mirroring
- How it works: Data is duplicated across two drives
- Pros: Simple, fast read speeds, survives one drive failure
- Cons: 50% usable capacity, write speed limited to single drive
- Best for: OS drives, small critical datasets
RAID 5 — Striping with Parity
- How it works: Data and parity information distributed across 3+ drives
- Pros: Good balance of performance, capacity, and redundancy
- Cons: Slow rebuild times, vulnerable during rebuild (RAID 5 write hole)
- Best for: File servers, NAS, low-write workloads
RAID 6 — Double Parity
- How it works: Like RAID 5 but tolerates two simultaneous drive failures
- Pros: Higher fault tolerance, safer for large-capacity drives
- Cons: Slower write performance, requires minimum 4 drives
- Best for: Production databases, archival storage, large HDD arrays
RAID 10 — Mirroring + Striping
- How it works: Combines RAID 1 mirroring with RAID 0 striping
- Pros: Excellent performance and redundancy, fast rebuild
- Cons: 50% usable capacity, requires minimum 4 drives
- Best for: High-transaction databases (MySQL, Oracle), virtualization hosts
How to Choose a RAID Controller
Key Specifications to Compare
| Spec | Entry-Level | Enterprise-Grade |
|---|---|---|
| Interface | PCIe 3.0 x8 | PCIe 4.0 x8/x16 |
| Port Count | 8 ports (SAS/SATA) | 16–24 ports (SAS/NVMe) |
| Cache | 1GB DRAM (no BBU) | 4–8GB DRAM + CacheVault/BBU |
| Protocol | SAS 12Gb/s | SAS 24Gb/s / NVMe |
| Example Models | LSI MegaRAID 9460-8i | Broadcom MegaRAID 9670W-16i |
Critical Features for Production Environments
- Battery Backup Unit (BBU) or CacheVault: Protects cached writes during power loss — essential for write-heavy workloads
- RAID-on-Chip (ROC): Dedicated hardware processor offloads RAID computation from the host CPU
- Hot-spare support: Automatically rebuilds a failed drive using a standby spare
- Vendor HCL listing: Ensure the controller is on your server vendor’s Hardware Compatibility List
Common Mistakes
- ❌ Using RAID 5 with large-capacity HDDs (8TB+) — rebuild times can exceed 24 hours, risking a second failure
- ❌ Skipping BBU/CacheVault on write-intensive workloads — risks data corruption on power loss
- ❌ Confusing software RAID (mdadm) with hardware RAID — software RAID consumes CPU resources and lacks cache protection
- ✅ Always test RAID rebuild procedures before going to production
Conclusion
Selecting the right RAID controller and level is a balance between performance, redundancy, and cost. For most production environments, RAID 6 or RAID 10 with a hardware controller featuring BBU/CacheVault is the recommended baseline. If you need help selecting a bench-tested RAID controller for your specific server platform, contact our team for expert guidance and competitive pricing.
Matching the RAID level to the failure tolerance you need
- RAID 1 and 10 tolerate a single drive failure per group; RAID 5 tolerates one per array; RAID 6 tolerates two
- Availability requirements matter more than raw capacity: a two-drive tolerance is often worth the extra parity
- In virtualised estates, think about how a degraded array behaves before the rebuild finishes
Capacity and the usable-versus-raw maths
- RAID 5 costs one drive of capacity; RAID 6 costs two; RAID 10 costs half the array
- Capacity pressure pushes towards RAID 5 or 6, and that is exactly when rebuild time gets risky
- Running an array near full makes rebuild behaviour worse, so leave headroom
Rebuild time and why it drives the choice
- Rebuild time scales with array size and inverse drive speed — large near-full arrays are the risk case
- Rebuild competes with foreground I/O, so a slow rebuild on a busy array hurts the workload
- Cache and battery protection on the controller shorten rebuild and protect the cache
Related reading
- Data center RAID strategy — controller selection alongside level selection
- RAID vs HBA — the software-managed alternative
- MegaRAID 9560-16i — a tri-mode controller with cache protection
- RAID controllers — category hub
Frequently Asked Questions
Which RAID level should I choose for a server?
Start from the failure tolerance you need and work backwards. RAID 1 or 10 for small critical sets where rebuild must be fast; RAID 5 for large read-heavy sets where one-drive tolerance is acceptable; RAID 6 where you cannot tolerate a second failure during a rebuild. If you are unsure, RAID 10 with a mirrored pair is the safe default for a modest array, and it rebuilds faster than a wide parity set.
How much usable space do I lose to RAID 5 or RAID 6?
RAID 5 costs one drive of raw capacity, RAID 6 costs two, and RAID 10 costs half the array. That is the arithmetic; the harder question is what it means at rebuild time. A large array running near capacity has a long rebuild, and during that window it is vulnerable. Leaving roughly 20 percent headroom keeps rebuild time manageable.
Is RAID 10 better than RAID 5?
For write-heavy and latency-sensitive workloads, yes — RAID 10 gives near-native write performance and much faster rebuilds, at the cost of half your capacity. For large read-heavy arrays where capacity is the constraint, RAID 5 remains more efficient. The honest comparison is throughput and rebuild time against capacity, and there is no universally correct answer.
How long does a RAID rebuild take?
It depends on array size, drive speed and controller rebuild performance, and it scales badly with capacity. A wide, near-full array on slow drives can take a day or more, and during that time the array is degraded. Controller cache and battery-protected write cache shorten the rebuild and keep the array responsive while it runs.
Does RAID protect against a controller failure?
No. RAID lives on the controller, so a controller failure takes the whole array with it. That is the argument for redundancy at the disk level and for treating the controller as a single point of failure — and it is why some estates run an HBA with software RAID instead.
How do you verify a used RAID controller before shipping it?
We clean, re-flash and bench-test on our own test line, then verify the part number, cache size, connector count, port mapping, firmware level and the CacheVault supercap status. We report the supercap state rather than leaving it to be discovered at install time. Ask for photos of the actual card before you commit.