"We need more storage" is the least useful requirement in infrastructure, because it hides four different problems that need four different machines. Before you compare products, work out which of these you actually have — the answer usually makes the hardware decision for you.
Four workloads that all sound like "storage"
Backup target. Write-heavy in bursts, read almost never, until the day you read all of it at once. What matters is capacity per euro, write throughput during the backup window, and restore throughput on the worst day of your year.
Media library or object store. Read-heavy, concurrent, latency-sensitive per object but not per byte. What matters is IOPS and network egress.
Cold archive. Written once, kept for years, read on legal or audit request. What matters is capacity per euro and nothing else. Do not buy NVMe for this.
Working set that outgrew the boot disk. A database, a build cache, an application's uploads directory. What matters is latency and IOPS, and the capacity number is almost incidental.
The first three are storage problems. The fourth usually is not — it is a performance problem wearing a capacity costume, and buying bulk storage for it makes things worse.
The dividing line, in numbers
The honest way to choose between a storage VPS and a dedicated storage server is to look at where each one runs out.
NexonHost's storage VPS range runs from 200 GB to 1.6 TB of SSD:
| Plan | vCPU | RAM | Storage | Bandwidth |
|---|---|---|---|---|
| STG1 | 1 | 2 GB | 200 GB SSD | 4 TB |
| STG2 | 2 | 4 GB | 400 GB SSD | 8 TB |
| STG3 | 4 | 8 GB | 800 GB SSD | 16 TB |
| STG4 | 8 | 16 GB | 1.6 TB SSD | 32 TB |
Every tier ships a 10 Gbps port and DDoS protection.
Dedicated storage servers start roughly where that range ends and go to 200 TB and beyond — the entry build pairs two SSDs with an eight-bay SATA array of 10–20 TB drives; denser builds run 18 NVMe drives or 36 SATA drives, with RAID 0/1/5/10 selectable at checkout and unmetered 1 Gbps ports that scale to 10 or 20 Gbps depending on the build.
So the line is not subtle:
- Under ~1.5 TB — storage VPS. You are nowhere near needing a chassis full of drives, and virtualised SSD will outperform bulk HDD on almost every access pattern.
- Over ~2 TB — dedicated storage server, and the question becomes which drive type and which RAID level.
- Between the two — decide on I/O pattern rather than capacity. Random access says SSD/VPS; sequential bulk says HDD/dedicated.
Raw capacity is not usable capacity
The single most common sizing error is quoting the raw number. Eight 10 TB drives is 80 TB raw and never 80 TB usable:
| Layout | Usable from 8 × 10 TB | Survives |
|---|---|---|
| RAID 0 | 80 TB | nothing |
| RAID 5 | 70 TB | 1 drive |
| RAID 6 | 60 TB | 2 drives |
| RAID 10 | 40 TB | 1 per mirror pair |
Then subtract filesystem overhead, and leave headroom — a copy-on-write filesystem that passes about 80% utilisation behaves very differently from one at 50%.
A realistic planning figure for an eight-bay array of 10 TB drives in a redundant layout is 50–60 TB usable, not 80. Size against that number.
One warning that matters more as drives get bigger: RAID 5 on large drives is a poor bet. A rebuild after a failure reads every remaining drive in full, which takes many hours on 10 TB+ disks, and during that window a second failure or an unrecoverable read error loses the array. RAID 6 or RAID 10 is the safer choice at this capacity — our guide to RAID levels works through why. The checkout offers RAID 0/1/5/10; if you want a different layout or a denser array, the storage server page has a custom quote path, and that is the conversation to have before ordering rather than after.
And the line that has to be in every storage article: RAID is not a backup. It survives drive failure. It does not survive rm -rf, ransomware, a corrupt write, or a fire.
Size the bandwidth, not just the disk
This is the step people skip, and it is where storage plans actually get chosen wrong.
A storage VPS ships a metered allowance — 4 TB on STG1 up to 32 TB on STG4. That is generous for a backup target and tight for a media server. Do the arithmetic before you buy:
- Backup target, 200 GB dataset, daily incrementals plus a monthly full: perhaps 400–600 GB of writes a month. Trivial against 4 TB. But a full restore is another 200 GB, and a disaster recovery test that you actually run quarterly is another. Still comfortable.
- Media library serving 5 TB of video to real users: egress is the entire cost model, and it will exceed the allowance long before the disk fills.
Dedicated storage servers ship unmetered ports, which is why bulk-egress workloads end up there even when the capacity would fit on a VPS. If your monthly transfer is a multiple of your stored data, the port matters more than the disk.
A decision procedure
- Name the workload — backup, media, archive, or working set. If it is a working set, stop; buy IOPS, not capacity.
- Measure the dataset today and its growth rate. Twelve months of growth, not current size.
- Pick the redundancy layout, then compute usable capacity. Plan to 50–60% of raw for redundant layouts on large drives.
- Compute monthly transfer, including restores and DR tests. Compare against the allowance, or conclude that you need an unmetered port.
- Decide the recovery objective. How long may a full restore take? That number sets your minimum read throughput and often your network port.
- Then choose the product.
Steps 1 to 5 take an afternoon and save more money than any amount of price comparison.
Common mistakes
- Buying capacity for an IOPS problem. A slow database does not get faster on a bigger disk.
- Sizing to today's dataset. Storage migrations are painful; buy twelve months of headroom.
- Forgetting the restore. Backups you cannot restore inside your recovery window are not backups.
- Treating RAID as redundancy of last resort. One copy in one building is one copy.
- Ignoring the port. A dense array behind a saturated link is a slow array.
Where to start
If the dataset is small and the access pattern is random, start on a storage VPS and scale the tier as it grows. If it is large, sequential, or egress-heavy, go straight to a dedicated storage server — the entry Intel Xeon build provisions automatically, and the denser NVMe and SATA builds are quoted per order.
Both sit alongside the rest of the dedicated server range in twelve locations, so a backup target in a different country from the primary — which is the point of a backup target — is one order rather than a second vendor.
Related reading: understanding RAID levels and VPS vs dedicated server: which one should you choose.
