Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Backup, replication, failover and disaster recovery: four terms a buyer should not confuse

Technical article
Backup & Recovery

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.

These four terms answer four different questions. A backup answers: can I get yesterday's data back? Replication answers: is there a second copy that is almost current? Failover answers: can the service switch to that second copy quickly? Disaster recovery answers: what is the whole plan, and who runs it, when something large goes wrong?

A quote that says it includes disaster recovery may therefore mean anything from nightly file copies to a standby environment that takes over in minutes. The two are not interchangeable, and they do not cost the same. This guide gives a plain definition of each, shows which failures each one does and does not cover, and ends with a way to map your own systems so you can compare quotes on what they protect rather than on the label.

Why it matters

The common mistake is to assume that a service which copies data somewhere else also protects you from every way data can go wrong. It does not. Standards bodies and cloud vendors say so directly.

The US National Institute of Standards and Technology (NIST) warns in its contingency planning guide that "HA systems cannot be a replacement for a solid backup strategy, as a corruption of data on a system may propagate through an HA system, making the system unusable." (NIST SP 800-34 Rev. 1, section 5.1.6.) HA means high availability: redundancy built in so a system stays up. The same idea applies to replication. A copy that follows the original faithfully will also follow a mistake faithfully.

Amazon Web Services (AWS) makes the same point about replication in its disaster recovery whitepaper: continuous replication "may not protect you against data corruption or destruction unless your strategy also includes versioning of stored data or options for point-in-time recovery." (AWS, Disaster recovery options in the cloud.)

In our reading, this is why a buyer should ask which failure a service answers, not which product name it carries.

The four terms in plain English

Vendors define these slightly differently, so the definitions below are our summary of how the sources use them, not a single official wording.

Backup. NIST's glossary describes a backup as "A copy of files and programs made to facilitate recovery if necessary." The key properties are that it is a copy taken at a point in time and kept separately from the live system. Microsoft's reliability guidance says backups "involve taking a copy of your data and storing it safely for a defined period of time" and are useful "when automatic failover to another replica isn't possible, or when data corruption occurs."

Replication. Replication continuously copies changes from one system to another, so the second copy stays close to current. AWS describes a continuous, asynchronous form in which "versions of your data are available almost immediately in your DR Region." (Asynchronous means the copy follows the original with a short delay rather than waiting for confirmation.) Replication is about keeping a second copy fresh. It does not, on its own, keep old versions.

Failover. Failover is the act of switching the service to the secondary copy or site. Microsoft notes that "It usually takes some time for a failover process to detect that the primary instance has failed and to switch to the secondary instance." Failover needs a working secondary environment, not only a copy of the data. It also has a return trip: Microsoft says "Failback can be complex to plan and implement", because data written at the secondary site after the switch has to be dealt with when you go back.

Disaster recovery (DR). Microsoft defines it as "planning how to deal with uncommon risks and the catastrophic outages that can result." NIST's description of a disaster recovery plan is narrower: it "applies to major, usually physical disruptions" that deny access to the primary facility, and it "only addresses information system disruptions that require relocation." In plain terms, DR is a plan and a capability, not a single product. It typically uses backup, replication and failover as ingredients.

Two numbers that decide what you need

Two measures turn the vague question how protected are we? into something you can compare.

  • Recovery point objective (RPO): how much data loss is acceptable, measured in time. Microsoft's definition is "the maximum duration of acceptable data loss in the event of a disaster."
  • Recovery time objective (RTO): how long the service can be down. Microsoft defines it as "the maximum duration of acceptable downtime in the event of a disaster."

NIST uses the same two ideas under its own definitions, and notes that the point to which data can be recovered depends on "the most recent backup copy of the data."

Both are objectives you set, not results you are guaranteed. The RPO is a target for how much data you can lose. The RTO is a target for how long the service can be unavailable, and NIST ties it to selecting "appropriate technologies". Neither number is the same as what actually happens on the day.

Take daily backups. They give a potential recovery point of roughly 24 hours old, but only if the latest backup completed, can be read, is consistent and arrived on time. If the latest copy failed or holds the damage, you go back to an older clean copy and lose more than a day. The elapsed outage is also longer than the restore itself. It includes noticing and deciding, provisioning somewhere to restore to, the restore, checking the result and reopening the service. Microsoft says "Backup restoration often takes time", so a workload's RTO needs to account for it, and says testing lets you "validate the feasibility of your RTO". Microsoft also says "Because backups are typically taken infrequently, backup restoration usually involves data loss", and that backup recovery "should be used as a last resort." Replication can bring the potential recovery point down to seconds or minutes. Failover can bring the recovery time down to minutes. Each step down costs more, and Microsoft cautions that an RPO and RTO of zero is "difficult and costly to implement."

Which failure does each service cover?

The table is our illustration, built from the vendor and NIST statements above. Yes means the service is designed to help; Not alone means it does not cover that failure unless combined with something else. Your own design may differ, so use it as a prompt for questions, not as a promise about any product.

Failure Backup Replication Failover Disaster recovery plan
A file deleted by mistake last week Yes, if kept long enough Not alone: the deletion is copied across No Names who restores it
Ransomware encrypts live data Yes, if a clean copy is kept out of the attacker's reach Not alone: damaged data is copied across Not alone: it switches to a copy that may be damaged Sets the order of steps and decisions
One server or disk fails Yes, but slowly Yes Yes, if a standby exists Covers it if the failure is large
A site or datacentre is lost Yes, if the backup is held elsewhere Yes, if the copy is elsewhere Yes, if a secondary site exists Core purpose
A faulty software update corrupts data Yes, from before the update Not alone: the fault is copied Not alone Includes rollback and decision rules
The service must stay up through routine faults Not its job Supports it Core purpose Not the right tool

Two rows deserve emphasis. Ransomware and accidental deletion are two failures that replication is easily mistaken for covering. The AWS whitepaper recommends combining continuous replication with point-in-time backups for this reason, noting you "can back up the replicated data in the disaster Region to create point-in-time backups in that same Region." (Same AWS page as above.)

A second point concerns where the backup lives. AWS advises using a different account for a disaster recovery Region "in the case compromised credentials are part of your disaster recovery plans." Microsoft says backups used for DR "should be stored separately to the main data." Our reading: a backup that sits under the same administrator login as the live system is exposed to the same compromise.

Choosing the combination

The following is our recommendation, not vendor guidance.

  1. Start from the business process, not the server. AWS says a disaster recovery plan should sit inside the wider business continuity plan, not stand alone, and warns that recovery targets mean little "if that workload's business objectives cannot be achieved because of the disaster's impact on elements of your business other than your workload." Ask what the business must keep doing, then work down to systems.
  2. Set an RPO and RTO per system. A payroll system, a file share and a public website rarely need the same numbers. Microsoft says each component or flow "might have individual RPO and RTO values."
  3. Make backup the baseline. Almost every system needs a point-in-time copy kept separately, because it is the simplest answer to deletion, corruption and ransomware. It is not the only possible one. AWS says replication can also cover corruption when the strategy includes "versioning of stored data or options for point-in-time recovery", and a coordinated disaster recovery design can use that. Whichever route you choose, ask where the old, clean version comes from.
  4. Add replication where the RPO is shorter than your backup interval. Do it knowing it adds a fresh copy, not a history.
  5. Add failover only where the RTO justifies a standby. NIST observes that high availability "can be an expensive option for systems, with duplicate hardware and special failover software", and Microsoft advises: "Don't overengineer your solution to reach higher levels of reliability than are justified."
  6. Define the recovery requirements early, then keep the plan current and test it. Do not leave the plan until everything else is built. Record the dependencies, roles and decision points while you design, because Microsoft says the DR plan "will be used as part of infrastructure and application design". Write the runbook (the step-by-step recovery procedure) alongside the controls you select, and update it whenever they change. Then exercise it. Microsoft states that a DR plan should be "well-defined and testable", and that drills should "include all of the human processes in your drills, and don't just focus on the technical processes."

AWS describes four broad DR approaches that map onto this ladder: backup and restore, pilot light (a minimal always-on core), warm standby (a scaled-down working copy) and multi-site active/active (running in more than one place at once). AWS says backup and restore is "a suitable approach for mitigating against data loss or corruption." More active approaches cost more and recover faster.

Questions to put to your provider

Use these when a quote says backup, replication, failover or disaster recovery. Ask for the answers in writing.

  • Which of the four does this line item deliver? Ask for the word that describes the mechanism, not only the marketing label.
  • How often is a copy taken, and how long are old copies kept? This sets the RPO and how far back you can go after corruption.
  • Where does the copy live, and under whose credentials? Ask whether one compromised administrator login can reach both live data and copies.
  • What is the stated recovery time, and for which systems? Ask whether it has been measured or is a design estimate.
  • Is there a standby environment, or only data? Data without servers, networking and configuration is not a failover.
  • Who decides to fail over, and who decides to fail back? Ask who calls it and how long the decision usually takes.
  • When was a recovery last tested, and what did it show? Ask to see the result.
  • What is excluded? Ask which failures the service explicitly does not cover.

A worked example (illustrative)

The following is invented. A small firm runs an accounts system, a file server and a public website. Its supplier's quote lists backup and DR as one line. The firm's owner works through the matrix.

For the accounts system, a day of lost entries is unacceptable but two hours of downtime is tolerable, so the owner asks for backups several times a day plus a tested restore. For the file server, overnight backups and a day to restore are acceptable. For the website, a few hours of downtime costs little, so it needs backup and a written rebuild procedure rather than a standby. The owner asks one question of the quote: which of the four is this? It turns out to be nightly backup only. The firm now knows what to ask for, and what it is not getting.

Risks and limits

  • The terms are used inconsistently across vendors. NIST itself notes that "universally accepted definitions" for these planning areas "have not been available." Treat the definitions here as a working vocabulary, and check what each quote means by its own terms.
  • The failure table is our illustration. A particular product may cover more or less than shown.
  • Vendor documents describe their own platforms. The AWS and Microsoft pages are used here for definitions and design principles, not as a comparison of providers.
  • This guide does not test any product. A stated recovery time is a claim until a recovery has been timed.

Next step

List your critical systems, then for each one write down the failures from the table and which of the four answers it. If you are comparing quotes, the Backup and Recovery page is a starting point for that conversation; ask for the scope in writing.

Sources

All accessed 15 November 2026.

Learning notes

  • A backup is a separate copy from a point in time; replication is a copy that keeps up with the original, mistakes included.
  • Failover needs a working standby, not just data, and switching back afterwards is its own task.
  • Disaster recovery is the plan that uses the other three. It is not a product you can switch on, and it is defined during design, not added at the end.
  • Two targets, how much data you can lose and how long you can be down, let you compare quotes fairly. They are objectives, so ask whether a real recovery has been timed against them.