About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
In Veeam's documentation, "forever forward incremental" means one full file plus incremental files, with the oldest increment merged into the full once retention expires so the chain does not grow without limit. Other products use similar words differently: Bareos's always incremental consolidates older incrementals into a new backup instead. So the label describes how data is collected and stored in one product's terms. It does not tell you how fast a restore is, how many files a restore must read, or whether anyone reads the chain back to prove it works.
The same is true of "synthetic full", "image" and "snapshot". Each word has a documented meaning in a specific product, and a different meaning elsewhere. Read the label as a question to ask, not an answer.
What does each label mean in the vendors' documentation?
The table uses the vendors' own documentation. Where products differ, both are shown.
| Label | What the documentation says | What a restore needs | Ask |
|---|---|---|---|
| Incremental | Stores what changed since the previous backup (Bareos: files changed since the last successful backup of the same job) | The last full and every incremental up to the chosen point | How many incrementals can sit on one full? |
| Differential | Bareos: everything changed since the last full backup, which "effectively merges all the Incremental and Differential backups since the last Full backup into a single Differential backup" | The last full and the latest differential, plus any incrementals taken after that differential | How often is a new full taken? |
| Incremental forever (Veeam: forever forward incremental) | One full file plus incremental files. Veeam "injects data of an incremental file into the full backup file before deleting the increment" once retention expires | The full and the incrementals up to the point, though Veeam says the earliest point restores fastest | How are merged files compacted, and how often is the result checked? |
| Always incremental (Bareos) | Clients always run incrementals; older incrementals are consolidated into a new backup. Bareos limits it to file-based backups | The consolidated backup plus later incrementals | Is the backup file-based? Does it cover my VMs and databases? |
| Synthetic full | Veeam: a full synthesized on the repository from earlier data, "identical" in content to a regular full, with no read from the source | One full file, then the incrementals after it | Is the full made from stored data, and is that stored data verified? |
| Active full | Veeam: the source data is read again in full, "just as if you run the backup job for the first time" | Same as a full | How often, and at what load on my servers? |
| Image-level | Veeam: "copies a VM image as a whole at a block level", not files; used for whole-VM and file-level restore | Depends on the chain behind the image | Can I restore one file without restoring the whole disk? |
| Snapshot | See below; it has at least three meanings | See below | Where is the snapshot stored, and for how long? |
Why can "incremental forever" restore slowly?
An incremental stores changes only. To rebuild a given day, the software must combine the full with each increment up to that day. Veeam's health check page says the data blocks needed to "compose" the VM's latest state are "typically spread out across several backup files in the backup chain". The full and the newest increments are separate files. It follows that a longer chain means more files to combine; that last step is reasoning, not a sentence from the page.
Veeam also states the trade it makes. Merging increments into the full "can lead to the fragmentation of the full backup file", and a compact job is needed that "produces an additional I/O load on the backup storage". Its documentation says restoring the earliest point from this method is "the fastest compared to other methods", and that other points are comparable to the forward incremental method. Forward incremental, in contrast, adds periodic fulls: "By doing regular full backups, the backup chain is split into shorter series", which "lowers the chances of losing the backup chain completely".
Bareos describes the same problem from the other side. In its conventional scheme "jobs are removed from the system on which existing jobs depend", which its always incremental scheme avoids by consolidating.
What are the three meanings of "snapshot"?
A hypervisor snapshot. In VMware, a snapshot is not a backup. Broadcom's knowledge base states: "Do not use VMware snapshots as backups. The snapshot file is only a change log of the original virtual disk... If the base disks are deleted, the snapshot files alone are not sufficient to restore a virtual machine." It also advises not to keep a single snapshot "for more than 72 hours" and to delete snapshots after a successful backup. If a label says "daily snapshots" and these are hypervisor snapshots on the same storage as the server, they do not protect against loss of that storage.
A backup mode. In Proxmox VE, "snapshot" is a way of taking a backup while the VM runs. The documentation says Proxmox VE live backup "provides snapshot-like semantics on any storage type" and "does not require that the underlying storage supports snapshots". The mode name says nothing about where the copy lives.
A backup tool's restore point. In Proxmox Backup Server, each backup is a snapshot. Its documentation explains that "the actual upload of Snapshots is incremental while each Snapshot references all chunks and is thus a full backup". So the same word means a restore point that behaves like a full backup, built from an incremental upload.
Do these labels say anything about checking?
Not by themselves. A label describes how backups are made, not whether they are read back. Veeam's health check verifies "only the latest restore point in the backup chain". To run it periodically you "must enable the Perform backup files health check option"; its default schedule is 05:00 on the last Saturday of the month. If the latest point is incomplete, it checks the one before. Older points are not read.
For a forever-incremental chain that matters: a damaged block in an old increment can sit under every later point, and a single latest-point check may not touch it. Ask what is read back, how often, and how old the oldest checked point can be.
Example (illustrative): a provider offers "image-level, incremental forever, 30 daily snapshots" for fs01, which holds 2 TiB and gains about 40 GiB of changes a day. Read it as: one full copy of the disk at the block level (2,048 GiB), plus 29 increments of about 40 GiB (1,160 GiB), so roughly 3,208 GiB before compression and sharing of repeated data. The 30th-day point is the full plus all 29 increments combined. Restoring the oldest point needs the least combining; restoring yesterday needs the most. The questions then follow from the label: how are the oldest increments merged into the full, who checks the merged full, and when was a restore of yesterday's point last timed? The sizes are invented for the example.
How do I check this with my provider?
Ask for the answers in writing, in these terms.
- Name the method in product terms. Which product and version makes the backups, and which of its documented methods is it set to (for example forever forward incremental, forward incremental with synthetic fulls, or Bareos always incremental)?
- Chain length. What is the longest chain a restore must read? What forces a new full?
- Where the copy lives. Is it on separate storage from the server it protects? A label of "snapshot" can mean it is not.
- What is read back. Which points are verified, how often, and what happens when verification finds a problem?
- A restore you can see. When was a point restored, to where, and how long did it take? Ask to watch one.
- What cannot be restored this way. Bareos always incremental is file-based only; ask what covers the databases and virtual machines.
The related article on questions to ask a managed provider covers access, notice and exit. These six are about what the method label implies for a restore.
Risk: accepting a label without a timed restore leaves the restore time unknown. If a provider cannot show a restore of a point from the far end of the chain, treat the label as unproven and schedule a test restore before relying on it. There is no way to recover the time lost if the first restore is during an incident.
Read next
Related articles you may want next.
- Backup & RecoveryQuestions to ask a managed backup provider before you signQuestions to ask a managed backup provider about restores, access, incident notice, data location and exit, drawn from ASD outsourcing guidance.Read article
- Backup & RecoveryBackup, replication, failover and disaster recovery: four terms a buyer should not confuseBackup, replication, failover and disaster recovery cover different failures. See which protects against deletion, ransomware, outages and site loss.Read article
- Backup & RecoveryThe 3-2-1 backup rule in 2026: what each number must proveWhat the 3-2-1 backup rule says, what it misses, and how ASD's Essential Eight and ISM backup controls add access, immutability and testing to it.Read article
Risks and limits
- Each vendor's terms come from its own documentation for the version in the scope line. A provider using different software may use the same words differently.
- The documentation cited gives no restore times. The speed statements are the vendors' relative comparisons only.
- The example uses invented sizes and ignores compression and shared data.
- Label questions are about restore behaviour. Costs, retention length and storage that cannot be deleted are separate subjects.
- Vendors' pages may have changed since the dates shown.
NetLeverage includes a daily full disaster-recovery backup of each hosted server at no extra charge. Talk to an engineer.
Sources
- Backup Methods, Veeam Backup & Replication 13 User Guide — Veeam Software, page updated 2025-08-26, accessed 17 November 2026
- Forward Incremental Backup — Veeam Software, page updated 2026-03-11, accessed 17 November 2026
- Synthetic Full Backup — Veeam Software, page updated 2025-07-09, accessed 17 November 2026
- Active Full Backup — Veeam Software, page updated 2025-07-09, accessed 17 November 2026
- Health Check for Backup Files — Veeam Software, page updated 2026-08-04, accessed 17 November 2026
- Backup (image-level backups of VMs) — Veeam Software, page updated 2026-08-04, accessed 17 November 2026
- Director configuration (Level) — Bareos project, bareos-25, accessed 17 November 2026
- Always Incremental Backup Scheme — Bareos project, bareos-25, accessed 17 November 2026
- Backup and Restore (vzdump), Proxmox VE 9.2.14 — Proxmox Server Solutions, accessed 17 November 2026
- Technical Overview, Proxmox Backup 4.2.11-1 — Proxmox Server Solutions, accessed 17 November 2026
- Best practices for using VMware snapshots in the vSphere environment (KB 318825) — Broadcom, updated 11-15-2026, accessed 17 November 2026
This article was originally published on .