Articles

Full, incremental and differential backup

Three backup chains: what each costs to store, what a restore needs, and where they break.

Reading: 4 minServer & Virtualization

Article cover: Full, incremental and differential backup

A backup is a copy you can restore from; every job that makes one belongs to a chain. Three chains exist: full, incremental and differential. They differ in one decision — what each run copies — which sets both how much storage the chain occupies and how many pieces a restore must reassemble. The type changes what it costs to keep the data and to get it back.

The model: a base and the runs that hang off it

A full backup establishes a base. The later runs are defined against something, and what they are defined against decides both how the chain grows and what a restore must gather:

   base (full) ──▶ incr 1 ──▶ incr 2 ──▶ incr 3      incremental:
                                                      each run depends on the run before it
   base (full) ──▶ diff 1 ──▶ diff 2 ──▶ diff 3      differential:
                                                      each run depends on the base alone

A restore reads the base and then every run it does not already contain: an incremental chain is replayed link by link, while a differential chain needs the base and one run. NIST defines the full plainly — it “captures all files on the disk or within the folder selected for backup”. Its virtue is self-containment: because “all backed-up files are recorded to a single media or media set”, locating a version is direct. Its cost is in the same passage — a full run “can be lengthy”, and repeating fulls of unchanging files “could lead to excessive, unnecessary media storage requirements”.

Three ways to fill the chain

Chain What each run copies Depends on Size across the cycle Restore gathers
Full every selected file, or every used block nothing constant, the largest of the three the full alone
Incremental what changed since the last backup of any type the previous run in the chain small and roughly constant the full and every incremental since it
Differential what changed since the last full the last full only grows with each run until the next full the full and the last differential

The definitions carry the consequences. An incremental backup “captures files that were created or changed since the last backup, regardless of backup type” — its reference point is the previous run, whoever made it. A differential backup “stores files that were created or modified since the last full backup”, and it saves “the file each time until the next full backup is completed”.

NIST states both restore outcomes. If the last full was three days ago and one file changed each day, “the media for the full backup and for each day’s incremental backups would be needed to restore the entire directory”. For a differential, “Restoring from a differential backup may require fewer media than an incremental backup because only the full backup media and the last differential media would be needed” — and the price is stated without softening: “differential backups take longer to complete than incremental backups because the amount of data since the last full backup increases each day until the next full backup is executed.” In practice: “a full backup can be conducted on the weekend with differential backups conducted each evening”.

What a restore needs, and what it costs

The restore matrix is the summary: the full alone is self-contained and the slowest to produce; the incremental chain is the full plus every run since, as long to replay as to run; the differential chain is the full plus one run, so the intermediate differentials can expire early — a retention benefit, not a property of the chain.

SQL Server shows the names are not universal. Its differential backup is “based on the latest full backup” and contains “only the data extents that have changed since the differential base”; it has no type called incremental, its fine-grained chain being log backups, each of which “includes all log records that were not backed up in a previous log backup (full recovery model)”. Interpretation: that is the incremental idea applied to the transaction log rather than to files.

A differential is anchored to the full, so its size tracks the interval between fulls: skip a full and it drifts toward the size of one, and the restore weight with it. An incremental is anchored to the previous run, so each stays small however long the interval; what grows is the chain, and so the restore. That is the trade — the incremental minimises storage and backup time and maximises restore effort; the differential trades a heavier but bounded backup for a shorter restore. Which fits depends on what you can afford to lose and how long you can be down — recovery objectives for this area.

File-level and block-level are a different axis

The three names describe what a run compares: all of it, the changes since the last run, the changes since the full. Whether it compares files or blocks is a separate choice; block-level systems rediscover the same chains. An EBS snapshot is “an incremental backup” in AWS’s own wording, saving “only the blocks on the volume that have changed since the most recent snapshot” while still holding “the information that is needed to restore your data”. zfs send is the same at the stream level: “By default, a full stream is generated”, and -i produces “an incremental stream from the first snapshot (the incremental source) to the second snapshot (the incremental target)”. The unit changes what a change costs to detect, not the restore dependency.

A backup is only proven by a restore

A job that reports success has proved that data was written, not that it can be read back, joined and applied. CISA states the requirement in one line — “It is critical to maintain offline, encrypted backups of data” — and the practice: “regularly test your backups”. NIST adds the operational form: backups should be “routinely conducted, stored offsite, rotated, and periodically validated”, and media should be tested so that “data are being stored correctly and that the files may be retrieved”.

The only test that exercises a chain’s real dependency is a restore: the base plus the runs, in order, into an environment other than the one being protected. Restoring a full proves the full; restoring a chain proves the chain.

Where a chain breaks

  • A missing link. Every run after a lost incremental is unusable.
  • A full that never runs. Differentials keep growing until they approach a full.
  • A base expired too early. Retention policy, not the chain’s mechanics, decides how long the base survives.
  • A copy the attacker can reach. Offline and separate copies are what keep a chain restorable, since “many ransomware variants attempt to find and delete any accessible” backups.
  • A plan that never tests the restore. Frequency sets what you can lose; chain length sets how long a restore takes.

Level and prerequisites

L3 — chain mechanics without any product procedure: which run copies what, what depends on what, and what a restore therefore needs. Prerequisites: the block, file and object sheet; volumes, partitions and filesystems; RAID and backup — availability is not protection.

Where to go next

  • Server & Virtualization — the area this sheet belongs to.
  • Further sheets, named in prose: Snapshots and their limits (a copy, not a chain); Retention, immutability and separate copies (properties of a copy, not its mechanics); Replication and the mistakes it replicates; RPO and RTO.

References

  • NIST, Contingency Planning Guide for Federal Information Systems (SP 800-34 Rev. 1) — full, incremental and differential backup, and the recovery consequence of each.
  • Microsoft Learn, Backup overview (SQL Server) — differential base, full backup, log backup and the recovery model.
  • AWS, Amazon EBS snapshots — snapshots as incremental, block-level backups and their restore guarantee.
  • OpenZFS, zfs-send.8 — full and incremental send streams.
  • CISA / MS-ISAC, Ransomware Guide — offline, encrypted backups and the requirement to test them.
Nodrius Field Kit — cover

Get the Nodrius Field Kit

Enter your email and we'll send you the complete Field Kit: eight technical resources in one download.

Your email address is sent to Nodrius and used to deliver the Field Kit to you by email. Marketing messages are separate and optional: the checkbox does not affect your download, and you are only added to our updates list if you tick it. Privacy Policy.

Privacy & cookies

This site does not use cookies, analytics or tracking, and it does not profile you. There is nothing technical to switch off, so Accept and Reject change nothing about how the site works: either choice lets you browse everything normally.

The only difference is what your browser remembers: two entries in local storage — your choice here, so this notice is not shown again, and a marker that the Field Kit offer has already been made — so the offer is presented once and not on every page. They contain no identifier, they are not shared with anyone, and the site sets no cookie for either.

What we do with your data
How your email address is used when you ask for a Resource, how long it is kept and which rights you have — in the Privacy Policy.
What the site stores in your browser
Two entries in local storage — your choice on this notice and whether the Field Kit offer has been shown — and no cookies, no session storage, no third-party content: in the Cookie Policy.