Home›Guides›IT backup

IT backup

What is the difference between backup and replication?

Replication maintains a copy close to the current state, so you can resume quickly after a failure. A backup keeps earlier states, so you can go back to before an error or an attack. The two complement each other: one does not replace the other.

Updated October 20263 min read5 sources cited

Key points

  • Replication is an availability tool: it also copies deletions, corruption and encryption.
  • A backup is a rollback tool: it keeps several dates, at the cost of a slight delay (the RPO).
  • The ANSSI, France’s national cybersecurity agency, points to replication when you cannot lose more than a few hours, and to backup to return to a healthy state.
  • Microsoft 365 and Google Workspace recycle bins are short-term safety nets within the same tenant, not a backup.

Replication follows the original

A database replica, a virtual machine mirror or a file synchronisation sends changes to a second location, often within seconds. If the main server fails, you can switch to the second one and lose few transactions. It is an availability tool. The NIST contingency planning guide (SP 800-34) in fact reserves mirrored systems and disk replication for high-impact systems, combined with a standby site that is already running.

The replica also receives what should not be kept: a deleted file, a corrupted database, a document encrypted by ransomware. Depending on the mode (synchronous, asynchronous, with or without delay), the healthy copy disappears at the same time as the original, or a few minutes later.

A backup keeps the past

A 2 pm backup, a 6 pm backup and one from the previous evening remain available. If the attack started at 4 pm, you restore the 2 pm backup. You lose the afternoon’s work. You recover a usable system. This deliberate gap is the RPO.

A backup is generally less “fresh” than replication. It is the only one of the two that allows you to roll back. The ANSSI backup guide says so in its own way: it excludes from its scope requirements for data loss of less than 24 hours, for which it recommends synchronous or asynchronous replication.

Comparison

ReplicationBackup
PurposeResume quickly after a failureReturn to an earlier healthy state
Freshness of the copySeconds to minutesHours (depending on frequency)
HistoryNone or very shortSeveral days, weeks or months
Deletion, corruption, ransomwareCopied to the replicaEarlier versions intact
Outright hardware failureFast failoverRestore, which takes longer

How to combine them

NeedSuitable tool
Resume within minutes after a hardware failureReplication or BCP, with a second system already in place
Return to before a deletion or an attackBackup with history, ideally a copy the attacker cannot modify
BothReplication for outright failures, immutable backup for errors and ransomware

A company that does not replicate and backs up every night accepts redoing up to 24 hours of work, and waiting for the restore. A company that only replicates may restart quickly and discover that the replica is already encrypted.

The case of the cloud

A hosting provider’s geo-replication protects against a data centre fire. It copies the state of the volume, including a volume already encrypted by the attacker. It is not a backup history.

The Microsoft 365 and Google Workspace recycle bins are short-term safety nets, inside the same administrator account. In Exchange Online, a deleted item remains recoverable for 14 days by default, 30 days at most. In Gmail, an administrator has 25 additional days after the 30-day trash period; after that, neither the administrator nor Google can restore the message. These mechanisms are neither replication to a third party nor a backup outside the tenant. See Does Microsoft 365 really include a backup?.

At WeDoBack

WeDoBack backup keeps versions, at the frequency chosen by the client. The BCP is different: cloud instances run continuously and take over if a server goes down, without any change of IP address, which comes close to service continuity. Data replication or synchronisation between the BCP instance and the original server is not built in: it relies on a specific process, tailored to the need, which WeDoBack can set up on quotation. The DRP restarts servers on standby instances from a chosen backup version. In both cases, the backup history remains the way to return to a state before the incident.

Frequently asked questions

If I replicate my server to a second site, do I still need a backup?

Yes. The second site protects you from a hardware failure or a disaster at the first. It does not protect you from a deleted file, a corrupted database or ransomware: the replica receives these changes like any others. Only a backup history lets you go back to before the incident.

Is a OneDrive or Dropbox synchronisation a backup?

No. Synchronisation copies the current state in both directions: a file erased or encrypted on the computer is erased or encrypted in the cloud too. The service’s versions and recycle bin help with an isolated error, for a limited period, but they remain under the same accounts as production.

Does delayed replication protect against ransomware?

Only if the attack is detected before the delay runs out, often a few minutes or a few hours. Yet an intrusion usually remains undetected for several days. A replication delay does not replace several weeks of history.

Planning a backup, DRP or BCP project?

More than 20 years of experience protecting business data.

Request a quote+33 9 72 50 78 28

Protect your data with WeDoBack

Encrypted offsite backup, immutable storage, DRP and BCP: tell us about your servers and we will recommend the right combination.