RPO and RTO

RPO is how much data you can lose; RTO is how long you can be down

00:0002:0004:0006:0008:0010:0012:0014:0016:0018:0020:0022:0024:00nightly backupfailuredata lost: 12 hours of orders (the RPO you actually have)agreed RPO: lose at most 1 hour, so back up at least hourlyservice backdowntime: 4 hours (RTO)RPO looks back from the failure: how much data you can lose. RTO looks forward: how long until it works again.RPO is how much data you can afford to lose; RTO is how long you can afford to be down.The recovery point objective sets how often you back up or replicate: a one-hour RPO needs at least hourly copies.The recovery time objective sets how fast you must restore: minutes need standby systems, days allow a rebuild.Both come from the business, not from IT, and both must fit inside the maximum tolerable downtime (MTD).

The recovery point objective (RPO) sets backup frequency; the recovery time objective (RTO) sets restore speed; both sit inside the MTD.

Backup types

Incremental is quick to take and slow to restore; differential the reverse

SunMonTueWedThuFriSatGB written each nightIncrementalsince the last backupFull 500202020202020Saturday restore: full + 6 incrementals = 7 sets, all needed. Nightly: 20 GB.Differentialsince the last fullFull 50020406080100120Saturday restore: full + Saturday’s differential = 2 sets. Nightly: grows to 120 GB.Incremental is fastest each night and slowest to restore; differential is the reverse. Snapshots and replication add near-continuous copies.Incremental backups are quick to take and slow to restore; differential backups are the reverse.A full backup copies everything (500 GB here). Incrementals copy only what changed since the last backup of anykind, so a restore needs the full and every incremental since, and one bad set breaks the chain. Differentials copyeverything since the last full, so they grow each day but a restore needs only two sets.

Full, incremental and differential backups trade nightly size against restore effort; snapshots add frequent copies.

Backup and replication methods · Going deeper

Snapshots, replication, journaling and protecting the backups

Snapshots, replication and journaling shrink the RPO; backups need encryption, separate credentials and test restores.

  • Snapshot a point-in-time image of a disk or virtual machine, taken in seconds. Usually stored on the same system, so it is not a backup on its own.
  • Replication continuous copying to another site. Excellent RPO, but it also copies corruption and deletions, so it complements backups rather than replacing them.
  • Journaling records every change so a database can be rolled to any moment, such as one minute before an attack.
  • On-site and off-site on-site restores fast; off-site survives the loss of the building. Most designs use both.

Protecting the backups

Encryptbackups hold everything, so encrypt them and guard the keys
Separatedifferent admin accounts from production
Immutableobject lock or write-once storage
Testrestore real files on a schedule

The 3-2-1 rule

Keep one copy that attackers can’t reach, and prove it restores

3copies of the datalive data + two backups2different mediasuch as disk and cloud1copy off-sitesurvives fire or flood1offline or immutableransomware can’t alter it0errors on restoretested, not assumedIn 2024, 94% of organisations hit by ransomware said the attackers tried to compromise their backups (Sophos).Keep three copies on two media, one off-site, one that attackers cannot touch, and prove they restore.Ransomware crews hunt for backups first, deleting or encrypting them so you have to pay. An offline copy, or animmutable one that cannot be changed or deleted until its retention ends, survives even stolen admin rights.Backups need their own credentials, encryption and monitoring, and regular test restores.

The 3-2-1-1-0 rule: three copies, two media, one off-site, one offline or immutable, zero restore errors.

Availability and redundancy

Each extra nine cuts downtime tenfold and costs far more

AvailabilityDown per yearDown per monthWhat it usually takes99%3.65 days7.31 hoursone server, restored from backup99.9%8.77 hours43.8 minutesredundant parts, quick failover99.99%52.6 minutes4.4 minutesclustered across zones, automatic failover99.999%5.3 minutes0.4 minutesactive in several regions; very costlyRedundancy removes single points of failure: RAID disks, dual power supplies, clusters, load balancers, multiple zones.RAID is not a backup: it copies a deletion or an encryption to every disk instantly.Each extra nine of availability cuts downtime tenfold, and costs far more than the last.Availability is the share of time a service works. Redundancy keeps it running through a failed part: a mirroreddisk, a second power supply, another server behind the load balancer, another data centre. Choose the targetfrom what downtime costs the business, not from what sounds impressive.

Availability targets (99.9%, 99.99%) need redundancy: RAID, clusters, load balancers and multiple zones. RAID is not a backup.

Recovery sites

The faster you must recover, the more you keep running in advance

SiteHas readyRunning again inCostCold sitespace, power and network onlyweeks$Warm sitehardware and older copies of datahours to days$$Hot sitea full duplicate with live dataminutes$$$$Cloud regionstandby you pay for only when usedminutes to hours$ to $$$Far enough away that one flood, fire or power cut cannot take out both sites.A recovery site trades cost against speed: the faster you must be back, the more you keep running in advance.A cold site is an empty room you fill after the disaster. A warm site has equipment and recent-ish data. A hotsite mirrors production and can take over in minutes. Cloud makes a hot or warm standby affordable, becausemost of the capacity is only paid for when you fail over.

Cold, warm and hot sites, and cloud standby, trade cost against recovery time; keep geographic dispersion.

Business impact analysis

The business, not IT, decides what comes back first

Business processMTDRTORPORecovery orderCard payments4 hours1 hournear zero1Customer website24 hours4 hours1 hour2Payroll3 days24 hours24 hours3Internal wiki2 weeks1 week24 hours4The RTO must be shorter than the MTD, with room for detection and decisions.BCP keeps the business working (people, sites, manual workarounds); DRP restores the technology.A business impact analysis decides what to recover first, and how fast, by what each hour of outage costs.The BIA asks each process owner: what breaks, and when does it become intolerable? That is the maximum tolerabledowntime (MTD). RTO and RPO are set inside it, which then decides backup frequency, sites and spend.The business continuity plan and the disaster recovery plan are both built on it.

A business impact analysis sets MTD, RTO and RPO per process and drives the BCP and the DRP.

Continuity planning · Going deeper

BCP, DRP, COOP and the metrics around them

Know BCP, DRP, MTBF, MTTR, succession planning and failover.

  • Business continuity plan (BCP) keeps essential functions going during a disruption: alternate locations, manual workarounds, communication, people.
  • Disaster recovery plan (DRP) restores IT systems and data after a disaster, in the order the BIA sets.
  • Continuity of operations (COOP) the government term for keeping essential functions running.
  • Succession planning who decides if key people are unavailable.

Reliability metrics

MTBFmean time between failures: how reliable
MTTRmean time to repair: how fast you fix
MTDmaximum tolerable downtime: the hard limit
RTO, RPOthe recovery targets set inside the MTD

Power counts too: an uninterruptible power supply (UPS) bridges seconds to minutes; generators run for days with fuel contracts in place.

Testing recovery

A recovery plan nobody has tested is only a hope

Tabletop exercisetalk through a scenario around a tableWalkthrougheach team steps through its part of the planSimulationact out a realistic scenario, without touching productionParallel testbring up the recovery site alongside productionFull interruptionactually fail over production: realistic and riskymore realistic, more disruptive, more costly ↑A recovery plan that has never been tested is a hope; tests run from talking it through to failing over for real.Tabletop exercises are cheap and find gaps in roles and decisions. Parallel tests prove the technology workswithout risk to production. Full interruption tests prove everything, at real risk. Every test ends withlessons learned and an updated plan; a test restore of backups belongs on the calendar every month.

Test with tabletop exercises, walkthroughs, simulations, parallel tests and full interruption tests.

Lesson complete

You finished “Backup and recovery”

Nicely done.

Open in a new tab