דילוג לתוכן
Articles and Knowledge Centre / Backup and Recovery

Types of backup: full, incremental, differential, and synthetic

Full, incremental, differential, synthetic — sometimes it seems we also need to back up the glossary. In practice, the difference starts with a simple question: do we save all the data or only what has changed? Let's clarify the types, recovery methods, and considerations for selection. If you are already familiar with…

12.10.2026 · 14-minute read
Illustration of a database backup, restore points, and backup sequence

Full, incremental, differential, synthetic — sometimes it seems you also need to back up the list of terms. In practice, the difference starts with a simple question: do you keep all the information or only what has changed? Let's organize the types, by restoration method and considerations for choosing. If you already know the basics, you can move on to the comparison and selection diagram that comes later.

The key principles

  • Full is the baseline
  • Changes are saved in different methods
  • Also consider recovery time
Key principles at a glance

Full backup: starting from all data sets

A full backup saves the entire data set you selected and transfers it to the backup destination. It is the foundation for the change methods in this guide. "Full" refers to the chosen set — not necessarily everything that exists in the organization.

If you copy everything in every run, a large portion of the information may be transferred again even though it hasn’t changed. To save transfer and storage, you can save changes with a incremental or differential backup, or build a synthetic full point. You don't have to send the entire directory again because of one page that changed.

incremental backup: only what has changed since the last time

An incremental backup saves what has changed since the previous backup. Each run adds the changes since the last point. This allows reducing transfer and storage use, and sometimes backing up more frequently.

How does the chain work?

The incremental backup creates a chain. As it lengthens, in the configuration described in the source, more time and resources may be required to check the full backup and changes and to compare between the computer and the destination.

The source recommends creating a periodic full backup and starting a new sequence. Its frequency can be weekly, monthly, or once every few months, as needed.

In a dependent chain configuration, it is impossible to delete a link that later restorations still need. The source describes deleting the entire series, from the full base to the next full backup, and recommends creating periodic full bases.

Types of incremental backup

Methods can be distinguished according to the size of the change unit that the system saves. The guide focuses on two common methods.

File-level incremental backup

In file-level incremental backup, a change to a file causes the entire file to be sent as a new version. This is simple and may be suitable for a small set of data. In a large file, even a small change can require a lot of time and space.

Block-level incremental backup

In block-level backup, the system checks parts within the file and copies only the blocks that have changed. Under suitable conditions, this can reduce transfer compared to sending the entire file. The unit that changes is smaller, so the work may also be smaller.

Points emphasized by the source

  • Block-level backup can reduce time and storage volume.
  • Backing up the entire file on every change is less efficient for large files that change frequently, because the file is sent again.
  • Restoring from a blockchain may be slow: the software needs to read the base and the required changes, assemble the file, and return it.
  • When a change missing in the chain is dependent, it can affect restorations that require it. Therefore, the source recommends occasionally starting a new chain with a full backup.

Differential backup: the changes since the full base

A differential backup stores new information and changes since the full backup it is based on. Here, the reference point is not the last backup, but the full base. Therefore, the difference from a incremental backup is not just in the name — it also changes which copies are needed for restoration.

The source emphasizes that it is usually faster than a full backup, and that its size grows over time. It describes using a differential backup in Microsoft SQL Server databases and other database applications, and presents it as less efficient in some file and system image scenarios.

Synthetic full: rebuilding from what has already been saved

Synthetic Full backup builds a new full point from information already stored in the repository and the new changes. Instead of reading and sending everything again from the source, it also uses what is already at the destination. This can reduce the load on the system being backed up.

How does it work?

After saving the first full base, changes continue to be saved. Occasionally, an additional full point is required, for example to support different recovery points, version retention, incremental chain at the block level, or in archive policy.

The source describes an implementation in which the system compares the data in the repository to the data on the computer, transfers only the changes, and copies the remaining information within the storage to create a new full set. In other implementations, a full backup is combined with incremental backups that already exist at the destination. The principle is the reuse of stored data, rather than a complete retransmission from the source.

Such an approach may be particularly efficient for large files and full system backups.

How was efficiency evaluated in the source?

The team described in the document conducted tests under different conditions, including with customers who participated in an early version test. Efficiency was measured using the ratio between data uploaded from the computer and data used in the storage.

  • Few changes and low fragmentation on the disk: In the “average user” scenario, a ratio of 20% from the computer and 80% from the cloud was described. When few changes were made, a ratio of 10% and 90% was also described.
  • Developer computers: With many changes and higher fragmentation, ratios of 60%/40% and 40%/60% were described.
  • Almost all data changed: In the least efficient scenario, a ratio of 92% local data and 8% from the cloud was described.

Additional points

  • The source indicates higher efficiency with large files.
  • In appropriate scenarios, a synthetic full can be faster than a regular full.
  • When there are few changes, a block-level incremental backup may still be faster than creating a synthetic full point.

How to choose without getting lost?

There is no backup type that wins in every situation. One examines what data is backed up, how much it changes, and how it needs to be restored. The following table summarizes the main difference between the methods:

Method What is saved? What is required for restoration?
Full The entire data set The corresponding full point
Incremental The change since the previous backup Base and the required changes according to the chain
Differential The change since the full base Base and the corresponding differential
Synthetic full A full point built from existing information and changes The outputs that the solution manager for this point produces

The original selection diagram, in an accessible view

The source diagram starts with a full base, then asks what type of information needs to be backed up. Here are its four paths in a format that can be read without deciphering arrows:

  • Part of the data: incremental backup at the block level.
  • System-level backup: full synthetic, and if not possible, incremental at the block level.
  • Microsoft SQL Server: full, differential, and transaction log, referred to in the diagram as "incremental."
  • Other databases or applications: incremental at the block level.

The source concludes with some rules of thumb:

  • Start from a full backup base.
  • Consider incremental backup for files.
  • Do not forget the management of full bases and the chain.
  • In image backup, examine the possibility of a synthetic full.
  • In SQL Server, full, differential, and transaction log backups are examined.

When choosing a method, check both the backup and the restore path. Fast transfer is an advantage, but on the day of a failure, you need to know which copies are required and how long the restoration takes.

Backup and restore in SQL Server

In a database, data backup and the transaction log are planned together. The recovery model determines which paths are available; a full backup is not a substitute for log backups when point-in-time recovery is required.

The process at a glance
08:00 · Full backup01
10:00 · Differential02
Logs up to the target time03
Educational example: restore full backup, the differential if taken, and then the required sequence of logs. Times are for illustration only.
  • Full: Contains the database data and log information required for backup consistency.
  • Differential: Collects changes since the corresponding full base backup.
  • Log: Maintains a sequence of transaction records and allows progression beyond the data backup, subject to the recovery model and correct sequence.

Keep copies separately from the database, restrict access, and perform a test restore to a separate database. Check integrity using appropriate tools, including DBCC CHECKDB after the restore, and document the result.

Source: Microsoft Learn — SQL Server Backup and Restore

Differential backup in SQL Server

The differential contains all changes since the full backup base, not just since the previous differential. To restore, the appropriate full backup and the desired differential are required; there is no need to apply all the intermediary differentials.

The process at a glance
Full · Base01
Accumulated changes02
Suitable latest differential03
Every differential relies on its full base. A Copy-only full backup does not change the differential base.

As changes accumulate from the base, the differential may grow. Schedule a new full base according to load and required restore time. Identify the base’s suitability from backup metadata, not just from the file name.

Source: Microsoft Learn — Differential Backups

Transaction log backup

Transaction log backups are supported in Full and Bulk-logged models, after the first full backup. They are not available in the Simple model. The backup frequency is derived from the amount of data that can be lost and the required recovery time.

The process at a glance
Full + differential is suitable01
Transaction log backup sequence02
Desired target time03
Required log in the chain must not be skipped. An additional full backup does not reset the transaction log backup sequence.

To restore to a point in time, check the sequence and backup that covers the target. In Bulk-logged, there are limitations when the log includes minimally logged operations. Before restoration, consult with the database administrator whether it is possible and necessary to preserve the Tail-log, to minimize the loss of recent changes.

Source: Microsoft Learn — Transaction Log Backup

Checked on 23.10.2026. The diagrams are original and intended for explanation.

Setting up a database backup in BackupONE

Before configuration, check the edition that supports SQL Server backup, permissions for the database and target, and the appropriate recovery model. Screen names vary by version.

From the database to the test restoration
SQL instance and databases01
Target, protection, and scheduling02
First backup and test restore03
Original workflow diagram, without customer data.
  1. 1

    Open a SQL plan

    In the component wizard, select MS SQL Server, local or cloud target, or a supported hybrid path, and give the plan a name.

  2. 2

    Select source

    Define SQL instance, authentication method, and permissions. Choose all databases, user databases, or a defined list.

  3. 3

    Protect and schedule

    Define compression and encryption, retention policy, full, differential, and log frequency according to the recovery model. Save the encryption key separately.

  4. 4

    Run and test

    Review the summary, run the first backup, and ensure the copy is available. Restore to a separate test database and validate the result.

Source: the manufacturer’s SQL backup configuration guide. This is an older source; verify the available options in the installed version. Checked on 23.10.2026.

Three models in SQL Server.
One choice that affects the path back to data.

Recovery models: what’s the difference?

The recovery model is a property of the database. It defines how the transaction log is managed, whether it can be backed up, and what recovery options are available to you.

Simple

No log backups. You revert to the end of an available full or differential backup; later changes are not covered.

Simpler management

Bulk-logged

Allows minimal logging for certain operations and requires log backups. When the log backup contains Bulk-logged changes, you cannot stop at a point within it.

Dedicated path for certain operations

Original concise processing in English based on Microsoft Learn — Recovery models.
The diagrams are illustrations and are not screenshots of the product.

Changing the model and starting the log chain

Before making a change, check the existing model and the backup plan. In SSMS, open the database properties, go to Options, and check the Recovery model field.

01Change from Simple
To Full or Bulk-logged
02Data backup
Full or differential
03Regular log backups
and restore testing
  1. document the way backBefore switching from Full or Bulk-logged, back up the log and ensure there is a documented recovery path.
  2. Start a log chainAfter switching from Simple to Full or Bulk-logged, take a full or differential data backup to start the log chain.
  3. Make the setting routineSchedule log backups, monitor free space, and update the recovery plan. Changing the setting alone does not create a backup.
  4. Complete the transitionAfter switching back from Bulk-logged to Full, back up the log. When switching to Simple, stop log backup jobs and adjust data backups.

Original concise processing in English based on Microsoft Learn — View or change recovery model.
Checked on 23.10.2026.

Continue reading