Dedicated server storage comes down to three choices: which drives, which RAID level, and whether the data belongs on that server at all. For most workloads the right answer is SSDs in RAID 1 or RAID 10 for the operating system and databases, big hard drives for bulk data, and backups on a different machine. Once your data outgrows the drive bays in one box, move it to a dedicated storage server or a small cluster.
Here’s how to make each of those calls without overpaying or painting yourself into a corner.
Shopping now? On our dedicated storage servers big drives are cheap: 14 TB enterprise HDDs are $79 a month each (8 TB $59, 4 TB $49; a 1 TB SSD is $59 and a 2 TB SSD $89), and a 10 Gbps private network between your servers is $20 a month. Chicago and Los Angeles builds are on the order forms; Ashburn is quoted. Or keep reading for how to pick the drives and the RAID level first.
Dedicated server storage options at a glance
| Drive type | Connects over | Strength | Weak spot | Good for |
|---|---|---|---|---|
| HDD (7,200 rpm) | SATA or SAS | Lowest cost per TB | Slow random I/O | Backups, media, archives, logs, object storage |
| SATA SSD | SATA 6 Gb/s | Fast random I/O, no moving parts | Capped by the SATA link | OS, web apps, most databases |
| NVMe SSD | PCIe | Highest throughput and parallelism | Cost per TB, fewer slots per server | Write-heavy databases, analytics, busy VM hosts |
HDD, SATA SSD or NVMe?
Hard drives are still the cheapest way to store a lot of data. Their problem is random access. A 7,200 rpm platter spins 120 times a second, so on average you wait about 4.2 ms for the right sector to come around, plus a few more milliseconds for the head to seek. Call it 8 to 12 ms per random read. That works out to roughly 80 to 120 random operations per second per drive. Sequential reads are fine, which is why HDDs are great for backups and video and miserable for a busy MySQL table.
SATA SSDs remove the seek entirely. Random I/O jumps by orders of magnitude, and for most web apps and databases a pair of SATA SSDs is all the speed you’ll ever use. The ceiling is the interface: SATA tops out at 6 Gb/s, which after encoding overhead is about 600 MB/s per drive.
NVMe drives skip SATA and talk to the CPU over PCI Express, with many parallel command queues instead of the single queue SATA drives use. That matters when dozens of threads hit storage at once: large databases with heavy writes, analytics jobs, or a hypervisor running a lot of VMs. If your application spends its time waiting on the network or the CPU, NVMe won’t make it faster. Measure first. iostat -x 5 showing drives near 100% utilization with long wait times is the signal that storage is the bottleneck.
What drives cost on our servers
These are the per-drive monthly add-on prices on GigeNET’s dedicated server order forms:
| Drive | Per month | Per TB per month |
|---|---|---|
| 4 TB HDD | $49 | ~$12.25 |
| 8 TB HDD | $59 | ~$7.38 |
| 14 TB HDD | $79 | ~$5.64 |
| 1 TB SSD | $59 | $59 |
| 2 TB SSD | $89 | $44.50 |
The gap is the whole story. A terabyte of SSD costs eight to ten times what a terabyte of 14 TB hard drive does, so put the hot data on SSD and everything else on spinning disks. Slot count matters too. Entry builds like the Xeon D-1541 take three drives. Dual Xeon E5 builds have eight drive slots, which leaves room for an SSD array and an HDD array in the same box.
Which RAID level should you use?
RAID spreads data across drives so a single dead drive doesn’t take the server down. Each level trades capacity, write speed and failure tolerance differently. Here’s how it plays out with four 14 TB drives:
| Level | Min drives | Usable from 4 × 14 TB | Survives | Use it for |
|---|---|---|---|---|
| RAID 1 | 2 | 14 TB (a 2-drive mirror) | 1 drive | OS and small databases on 2 or 3-slot servers |
| RAID 10 | 4 | 28 TB | 1 drive per mirror pair | Databases, VMs, anything write-heavy |
| RAID 5 | 3 | 42 TB | 1 drive | SSD arrays; avoid with large HDDs |
| RAID 6 | 4 | 28 TB | Any 2 drives | Bulk storage on large HDDs |
RAID 1 is the default for a small server. Two SSDs mirrored, done. On a three-slot build, a common layout is two SSDs in RAID 1 for the system and data, plus one large HDD for logs and local backup staging.
RAID 10 is what we’d pick for databases. You lose half the raw space, but writes are fast and rebuilds only copy one mirror, not the whole array.
RAID 5 looks like a bargain on paper and it’s the one we’d talk you out of on big hard drives. Every small write turns into two reads and two writes for the parity update, so it’s slow for databases. The bigger problem is rebuild time. Say a 14 TB drive fails and the array rebuilds at a steady 200 MB/s, which is optimistic on a server that’s still serving traffic. That’s 14,000,000 MB ÷ 200 MB/s = 70,000 seconds, nearly 20 hours of reading every surviving drive end to end. If a second drive fails or throws read errors in that window, the array is gone.
RAID 6 fixes that with a second parity block, so any two drives can fail. Writes are slower still, but for backups, media and archives that’s an easy trade. With eight 14 TB drives, RAID 6 gives you 84 TB usable.
Hardware RAID controllers and Linux software RAID (mdadm) both work well today. ZFS is a third route, with mirrors and RAIDZ2 as its equivalents of RAID 10 and RAID 6, and it adds checksums and snapshots. Pick one your team already knows how to rebuild at 3 a.m.
RAID is not backup
This one costs people their data every year.
RAID protects you from a dead drive. That’s it. If someone runs rm -rf in the wrong directory, if ransomware encrypts the filesystem, or if a buggy deploy corrupts your database, RAID faithfully copies the damage to every drive within milliseconds. It also does nothing if the whole server is lost.
A backup is a separate copy, on a separate machine, that the production server can’t overwrite. Every GigeNET dedicated server order form includes 1 TB of R1Soft backup space at $0 by default, which covers a first copy for a lot of servers. For an offsite copy, look at managed Veeam backups, or run your own with one of the tools in our open source backup software comparison.
When to move to a dedicated storage server
Keep data on the application server as long as it fits in the drive bays and only that server needs it. Move it to a dedicated storage server when one of these is true:
- You’ve run out of slots, or you’re about to.
- Storage is growing much faster than your need for CPU and RAM, and you don’t want to pay for compute you won’t use.
- Several servers need the same files.
- Your backups are sitting on the same machine they’re protecting.
The simplest version is one big server full of 14 TB drives in RAID 6 or ZFS RAIDZ2, sharing data over NFS or acting as a backup target. That’s a solid design up to the size of one chassis. Beyond that, or when you can’t tolerate the storage box itself going down, you want a cluster.
Our storage servers are dedicated servers sold in Ceph and MinIO builds. For the traffic between nodes, a 10 Gbps private network link is $20 a month on the order form, which keeps replication off your public port. If you’d rather not run storage yourself, see our storage services.
Network and object storage: Ceph and MinIO
Ceph gives you block, file and S3-compatible object storage from one cluster. By default it writes three copies of every object (osd_pool_default_size = 3), each on a different node, which is why three nodes is the practical minimum. Do the math before buying drives. Three nodes with three 14 TB drives each is 126 TB raw, about 42 TB usable with three-way replication, and you should plan to run well below that because Ceph gets unhappy as disks fill. Erasure coding stretches usable space further, at the cost of more nodes and more CPU.
MinIO is a lighter S3-compatible object store, popular as a backup target and for application uploads. One thing to know before you build on it: the open source MinIO repository is now marked as no longer maintained, and the community edition is distributed as source code only. MinIO points users to its AIStor products instead, which come under their own license terms. Read those terms before you commit. If you need S3 on hardware you control and want to stay fully open source, Ceph’s RADOS Gateway does the job inside a Ceph cluster. We compared the maintained options in MinIO alternatives in 2026.
Would rather not run object storage yourself? Our S3-compatible object storage runs in the same data centers as your dedicated server, with object lock for immutable backups. If it’s for training data or model files, see AI data storage.
Three setups that cover most cases
A web app with a database. An entry build with two SSDs in RAID 1 for the system and database, one large HDD for logs and local dumps, and the included R1Soft space for the first backup copy.
A database-heavy app. A dual Xeon E5 build with four SSDs in RAID 10 for the database and four 14 TB HDDs in RAID 6 for files and backup staging. Send the offsite copy somewhere else.
Tens or hundreds of terabytes, still growing. A Ceph cluster of at least three storage servers on a 10 Gbps private network, sized from usable capacity rather than raw.
Not sure which one fits? Send us your capacity, growth rate and read/write pattern and we’ll suggest a drive and RAID layout.
Not ready to order yet?
Tell us what you’re after and we’ll email you the build and the price. Need it today? Call (800) 561-2656 and we’ll check what’s in stock.
