Backup & storage
Keep the off-site copy of your 3-2-1 plan on a backup storage server you control: 64 TB of raw HDD, ECC memory and NVMe for metadata, in a different city from production.
Four 16 TB drives give 64 TB raw, or 32 TB usable with two-disk redundancy, at a far lower cost per terabyte than flash.
Storage servers run in New York, Los Angeles, Amsterdam and Frankfurt, so the copy can live in another city or on another continent.
Borg, restic, Kopia and Proxmox Backup Server can encrypt on the client, so the storage host only ever holds ciphertext.
Unmetered traffic means a full restore costs time, not money: 10 TB at 1 Gbps line rate takes about 22 hours.
Recommended servers
From a first deployment to a busy production setup. Live stock and prices; in-stock picks are online in minutes.
Our pickOff-site backup target
Ryzen 7 PRO 3700 · 8C / 16T · 3.6–4.4 GHz
4 × 16 TB HDD, 64 GB ECC and 2 × 512 GB NVMe for the OS and metadata: about 32 TB usable in RAIDZ2, in New York, Los Angeles, Amsterdam or Frankfurt.
Fast-restore tier
24C / 48T · 2.85–4.0 GHz
2 × 3.84 TB NVMe, 256 GB ECC and 10 Gbps for an all-flash PBS datastore or an S3 hot tier; add NVMe drives where a bay is free.
Large archives
EPYC 7313P · 16C / 32T · 3.0–3.7 GHz
12 × 20 TB HDD, 128 GB ECC and 10 Gbps for archives past 100 TB. Sold out: start on the 64 TB Storage Server, or request a custom build.
Less than 10 TB to protect? If you already run a DediJet server, off-server backup space starts at $5/mo, with no second machine to manage.
Sizing
Deduplication and compression ratios vary widely with the data (already-compressed video barely shrinks), so size from your own repository statistics after the first weeks.
| Backup scenario | Rule of thumb | Suggested plan |
|---|---|---|
| Borg or restic for 5–20 servers | Deduplicated history often takes 1.5–3× one full copy | 64 TB Storage Server |
| Proxmox Backup Server for a VM cluster | Dedup across guests; NVMe special vdev for metadata | 64 TB Storage Server |
| S3-compatible storage (Garage, SeaweedFS) | Parity RAID on HDD; NVMe for metadata and small objects | 64 TB Storage Server |
| ZFS replication target with syncoid | Pool larger than the source; snapshots kept longer | 64 TB Storage Server |
| Hot restores of VMs or databases | All-flash datastore and 10 Gbps to restore in hours | AMD EPYC 7443P |
| Media archive or long retention, 100 TB+ | 12 drives in RAIDZ2 vdevs, 10 Gbps for bulk moves | 240 TB Storage Server |
Expert notes
Short, practical advice from our engineers for this workload.
Four 16 TB drives can be arranged three ways:
Expect your tools to report 32 TB as about 29.1 TiB, and keep ZFS pools under roughly 80% full. The two 512 GB NVMe drives can hold the OS and, mirrored, a ZFS special vdev for metadata, which speeds up chunk-heavy repositories. Losing that vdev loses the pool, so never run it on a single drive.
The 3-2-1 rule asks for three copies on two kinds of media, one of them off-site. The bigger modern risk is an intruder with root on production who deletes the backups with credentials found there. Close that path:
borg serve --append-only with a forced command in authorized_keys, or run restic’s rest-server --append-only.MinIO, Garage and SeaweedFS all speak the S3 API, so restic, Kopia, rclone, pgBackRest and most backup tools can write to them unchanged. On a single server:
Root access means you install exactly what you need. These are proven choices for this workload.
FAQ
It has four 16 TB drives. In RAIDZ2, RAID 6 or as two mirrors you get about 32 TB usable, roughly 29 TiB in tool output; single parity gives 48 TB but survives only one failure. Keeping ZFS pools below about 80% full makes 25 TB a comfortable planning figure for a RAIDZ2 pool.
For a few servers’ worth of Borg or restic repositories, the backup space add-on is simpler: 1, 5 or 10 TB reachable over SFTP, rsync, BorgBackup or restic, separate from your server, starting at $5/mo. Choose a storage server when you need tens of terabytes, your own S3 endpoint, Proxmox Backup Server, or server-side snapshots and retention rules you control.
Yes. Boot the TrueNAS ISO from IPMI virtual media and install it as on any bare-metal machine; it then builds its ZFS pool on the four 16 TB drives. Keep its web interface off the public internet, behind a VPN or an IP allowlist, because storage appliances attract attackers. Proxmox Backup Server 4 and the main Linux distributions are one-click alternatives.
At line rate, 1 Gbps carries about 450 GB per hour, so 10 TB takes roughly 22 hours before disk and software limits. With the 10 Gbps upgrade, the network stops being the bottleneck and the HDD array usually becomes it. For recovery-time targets measured in hours, keep critical data on an NVMe tier such as an EPYC 7443P with 10 Gbps.
Split your repositories across two 64 TB Storage Servers, ideally in different locations, or pair them so each keeps a copy of the other’s critical data. If you need more than 100 TB in a single box now, request a custom build: it is quoted within a business day and usually ships in 2–5 business days. You can also join the restock alert for the 240 TB model.
Pick one of the recommended in-stock servers and it can be online in minutes. Need something bigger or different? Our engineers quote custom builds within one business day.