PBS Backup Architecture vs Traditional “Full / Incremental / Differential Backup”

Q: In PBS, I see many backup entries — which one is a full backup and which one is incremental?

This is one of the most common questions people have when they first start using Proxmox Backup Server (PBS):

Why are there so many backups, and why does each one look like a “full backup”?
Which one is the real full backup, and which ones are incremental?

The confusion comes from the fact that PBS works very differently from the traditional “full / incremental / differential” backup system.
Let’s break it down layer by layer 👇

🧩 1. PBS Backup Method: Deduplicated Snapshots (Not Traditional Full + Incremental)

Each backup you see in the GUI (e.g., ct/104/2025-10-23T06:02:57Z)
is logically a complete and restorable snapshot of the VM or container.
However, physically, PBS only stores the changed data blocks (chunks) compared to the previous version.

👉 In other words:

From the user’s perspective: every snapshot is a “full backup.”
From the storage perspective: only the changed chunks are new; unchanged data is shared from previous backups.

🔬 2. How It Works: Chunk-Based Deduplication

When PBS performs a backup, it splits the VM/LXC disk data into many chunks (default size: 4 MB each)
and calculates a SHA-256 hash for each chunk.

The process:

PBS starts backup and hashes each chunk.
If a hash already exists in the datastore (backed up before), it won’t be stored again.
Only new hashes (changed chunks) are saved.

So, even if you see multiple snapshots like:

ct/104/2025-10-22T10:03:00Z
ct/104/2025-10-22T12:02:55Z
ct/104/2025-10-22T14:09:22Z
...

What’s really happening is:

The first one (10:03Z) is the true initial full data set.
Each subsequent snapshot only adds the changed chunks,
while the index.json file still describes the entire VM.
During restore, PBS automatically reassembles all required chunks into a complete disk image.

📦 3. Example of Directory Structure

In a PBS datastore (e.g., /pbs/datastore1), each backup looks like this:

ct/104/2025-10-22T10:03:00Z/
 ├── index.json        ← Lists which chunks this backup uses
 ├── manifest.blob     ← Contains metadata (CPU, RAM, OS type, notes)
 ├── root.pxar         ← PXAR archive of the file system
 └── ...

But the actual chunk data shared across all backups resides in:

/pbs/datastore1/.chunks/

If a chunk appears in multiple backups, PBS stores it only once.

🧮 4. How to Identify the “First Full Backup”

Strictly speaking:

PBS does not label backups as full or incremental;
However, the earliest snapshot (the first timestamp) is the initial full backup;
All later backups are deduplicated snapshots (incremental in storage).

You can verify backup sizes via CLI:

proxmox-backup-manager snapshot list datastore1 --backup-id ct/104

Example output:

Snapshot	Logical Size	Stored Size	Note
2025-10-22T10:03Z	1.26 GiB	1.26 GiB	Initial full
2025-10-22T12:02Z	1.26 GiB	20 MiB	Incremental (changed chunks only)
2025-10-22T14:09Z	1.26 GiB	15 MiB	Incremental

📊 5. Checking in the GUI

In PBS GUI → Datastore → Content, enable the “Size” column (via top-right “Columns” menu):

If multiple snapshots have similar logical size but small stored size,
that indicates effective deduplication.
The datastore’s actual usage will not grow linearly with the number of backups.

✅ 6. Summary Comparison

Traditional Backup System	PBS Mechanism
Full + Incremental + Differential	Every snapshot is fully restorable
Incremental depends on previous backups	Each snapshot can be restored independently
High risk of space explosion	Deduplication + compression share chunks
Complex backup chains	Single datastore with automatic dedup management
Multiple versions take extra space	Identical data stored only once

💡 Practical Tips

✅ Frequent backups (e.g., every 2 hours) are fine — PBS only saves changed data, so space use is very efficient.
✅ You can safely delete old backups — PBS will automatically garbage-collect unused chunks.
✅ Set up a verify job to periodically check the integrity of each snapshot.