The three basic approaches
Backup software generally offers three ways of deciding what to copy. They differ in what they compare against, and that difference affects how long backups take, how much space they use and how complicated a restore becomes.
A full backup copies everything in scope, regardless of whether it changed.
An incremental backup copies only what changed since the previous backup of any kind, whether that was a full or an incremental.
A differential backup copies everything that changed since the last full backup. Each differential therefore grows larger as time passes since the full one.
Why the choice matters
A full backup is the simplest to reason about. To restore, you need that one set. The cost is time and space: copying everything every time can take hours and consume a great deal of storage, and it can load networks and disks heavily.
Incremental backups are quick and small, since each one holds only the latest changes. The price comes at restore time. You need the last full backup plus every incremental after it, applied in order. If any piece in the chain is missing or damaged, later pieces may be unusable.
Differentials sit in between. A restore needs just two parts: the last full backup and the latest differential. That is easier and less fragile than a long incremental chain, but each differential is larger than an incremental, and it keeps growing until the next full backup.
A hypothetical example
These numbers are made up purely for illustration. Suppose a file server holds 500 GB. A full backup runs every Sunday, and about 20 GB of different files change each day. Compare the Monday-to-Saturday backups.
Incremental: each day copies 20 GB. Over six days that is 6 x 20 = 120 GB, so the week totals 500 + 120 = 620 GB.
Differential: each day copies everything changed since Sunday, so the daily sizes are 20, 40, 60, 80, 100 and 120 GB. The six differentials add up to 420 GB, so the week totals 500 + 420 = 920 GB.
Now suppose the server fails on Friday evening, and you restore from the Friday backup (the fifth day after the full backup).
- Incremental: you need the Sunday full plus the Monday, Tuesday, Wednesday, Thursday and Friday incrementals, six pieces in all.
- Differential: you need the Sunday full plus the Friday differential, two pieces.
Real changes overlap: the same file is often edited on several days, so differentials would be smaller than in this tidy example. The pattern still holds. Incrementals use the least space and have the longest chain; differentials use more space and shorten the chain.
| Type | Backup speed | Space used | Restore effort |
|---|---|---|---|
| Full | Slowest | Largest | Simplest: one set |
| Incremental | Fastest | Smallest | Most involved: full plus every increment |
| Differential | In between | In between | Moderate: full plus latest differential |
Related terms you may see
Some products offer a “synthetic full” backup, which assembles a new full backup from an earlier full and later increments on the storage side, without recopying everything from the source. Others use “incremental forever,” keeping one initial full copy and merging changes behind the scenes. Many also use block-level or snapshot-based methods, which track changes at a finer level than whole files. The details vary by product, so read the documentation for whatever you use.
Choosing a combination
Most organizations do not pick just one. A common pattern is a full backup weekly or monthly, with incrementals or differentials in between. The right mix depends on a few questions:
- How much data changes? Heavily changing data makes frequent fulls expensive and differentials large.
- How quickly must you restore? Shorter acceptable downtime favors simpler restores. This is where recovery objectives come in, because they define the time and data loss you can tolerate.
- How much time is available for backups? Some systems have limited windows when copying will not slow users down.
- How reliable is your storage? Long incremental chains rely on every link surviving.
- How much space and bandwidth do you have? Sending data over the internet to cloud storage makes small incrementals attractive, though cloud costs and the time for a large initial upload need planning.
Protect the chain
Whatever combination you choose, keep a few habits. Verify backups after they run, because a corrupt link can make later pieces useless. Retain whole chains together, and do not delete an old full backup while newer increments depend on it. Keep at least one copy somewhere separate, following the idea behind the 3-2-1 rule. And practice restoring, since the true measure of a backup type is how smoothly you can get data back. A regular restore test will show whether your chosen method works as you expect and how long it really takes.