· AMBER team
How we size a Minecraft world backup
Why a 3 GB world can produce a 40 GB backup window, and the three settings that keep backups from eating the disk.
Minecraft worlds grow in a way that surprises people. A survival server with a 5,000-block border and a year of play can sit at 3 GB on disk. Turn on hourly backups with 21-day retention and the same server now needs tens of gigabytes of backup space — which is how a plan that looked generous on day one runs out of disk in month four.
Where the space goes
A region file is 512 by 512 blocks and is rewritten as players explore. Most of it is empty; the game allocates the full file anyway. A world with sparse exploration has a lot of nearly-empty region files, and each backup copies all of them whether they changed or not — unless the filesystem can snapshot.
On our nodes, backups are hardlink snapshots where the filesystem allows it. An unchanged region file is shared between snapshots, so fourteen daily backups of a mostly-static world cost close to one copy plus the changed files. That is why we can offer 14-day retention on Squad 4 without it being a lie on the disk side.
Three settings that matter
- Border. A 5,000-block radius is more world than a 20-player server will use in a year. Cutting it in half reduces the number of region files that get touched, which reduces the size of every backup.
- Backup frequency. Hourly backups make sense for a server where an hour of lost building is unacceptable. For most groups, daily plus a manual backup before a big build is enough.
- Entity cleanup. Dropped items and wandering mobs inflate chunk data. A periodic cleanup plugin reduces both server load and world size.
What we changed in the panel
Since 2.9.0 the backup view shows the size of each snapshot and the total space used, not just the dates. If retention is about to exceed the plan's disk, the panel warns two weeks ahead instead of failing silently when it fills.
Restores are unchanged: stop, copy, verify, swap. The one addition is that restoring now refuses to start if the target instance is running, instead of stopping it without saying so — a behaviour that produced one very confusing support ticket in 2024.
Backups on the same provider are not a disaster plan. Download a snapshot over SFTP every month and keep it somewhere else. We will not take offence; it is the correct thing to do.