Disaster Recovery in the Cloud: Strategies, RTO and RPO, and a Plan
Disaster Recovery in the Cloud: Strategies, RTO and RPO, and a Plan

Disaster recovery in the cloud means keeping copies of your data, and in some designs a ready-to-run copy of your systems, in a cloud provider’s infrastructure so you can restore service after an outage, a ransomware attack, a bad deployment or a regional failure. You choose how fast you need to be back (RTO) and how much data you can afford to lose (RPO), then pick the cheapest strategy that meets both. This guide walks through that decision in order, with the trade-offs and a plan you can test.
TL;DR
- Cloud DR replaces a second physical site with provider infrastructure, so you pay for variable usage instead of idle hardware.
- Set RTO and RPO per workload, based on what an hour of downtime or lost data actually costs. Tighter targets cost more.
- The four standard strategies are backup and restore, pilot light, warm standby and multi-site active-active. Cost and complexity rise as recovery time falls.
- A DR plan belongs inside your business continuity plan, and it only counts once you have restored from it in a test.
- Regulated data, such as patient records, adds documentation and testing duties on top of the technology.
Table of Contents
- What cloud disaster recovery is and how it works
- How cloud DR differs from traditional DR
- RTO and RPO: the two numbers that drive every decision
- The four cloud DR strategies
- Backups are not a DR plan
- What a cloud disaster recovery plan should contain
- Security, compliance and healthcare data
- Choosing an approach on a small-business budget
- Our perspective
- Work with CoreWorx on cloud recovery
- FAQ
- Sources
What cloud disaster recovery is and how it works
Disaster recovery (DR) is the part of business continuity that restores IT systems and data after a disruptive event. Google Cloud’s planning guide treats DR as a subset of business continuity planning that starts with a business impact analysis.
In the cloud version, the recovery environment lives with a provider instead of in a second office or colocation rack. In practice that means four moving parts:
- Replication or backup. Data, configuration and sometimes machine images are copied to cloud storage or a second region on a schedule or continuously.
- A recovery environment. Depending on the strategy, this is anything from empty (you build it during the incident) to a fully running duplicate.
- Failover. When the primary environment fails, traffic and workloads are pointed at the recovery environment.
- Failback. Once the primary is healthy, you move operations back and resynchronize data.
Microsoft describes the workflow as assessment, activation, failover, restoration, stakeholder notification, traffic redirection and service validation, and recommends scheduled drills so the steps are practiced before they are needed.
The disasters worth planning for are rarely dramatic. Most are a deleted production database, a compromised admin account, ransomware that encrypts both servers and attached backups, an expired certificate or a provider incident in one region. Tornado and storm risk matters for businesses in Oklahoma, but a single office losing power is a smaller problem once your systems no longer live in that office.
How cloud DR differs from traditional DR
Traditional DR usually means a secondary physical site, duplicate hardware and storage, or tapes moved off site. AWS’s DR guidance notes that the cloud shifts that fixed capital expense to variable operating expense and also points to simpler, more repeatable testing, lower management overhead and more automation.
| Traditional DR | Cloud DR | |
|---|---|---|
| Recovery site | Second building, colo or tapes | Provider region or zone |
| Cost model | Capital spend on hardware that may sit idle | Pay for storage and, depending on strategy, standby compute |
| Testing | Scheduled, disruptive, often skipped | Can be scripted and repeated on demand |
| Scaling at recovery time | Limited to what you bought | Capacity requested when needed |
| Main risk | Aging hardware, tapes nobody has read in years | Misconfiguration, region dependencies, surprise runtime bills |
The last row matters. A peer-reviewed comparison of cloud, colocation, on-premises and hybrid DR found that pay-per-use cloud recovery can save money while dormant but cost more while a recovery is actually running, and that compatibility, organizational constraints and workload traits can favor a hybrid design. The practical reading: cloud DR is usually the right default for a growing team, but you should model the cost of a multi-day recovery, not just the monthly standby bill.
RTO and RPO: the two numbers that drive every decision
Google Cloud defines RTO as the maximum acceptable length of time an application can be offline, and RPO as the maximum acceptable length of time during which data might be lost after a major incident. AWS words it slightly differently: RTO is the maximum acceptable delay between interruption and restoration, and RPO is the maximum acceptable time since the last recoverable data point.
A simple way to hold the difference:
- RTO is about the clock after the failure. If your order system dies at 10:00 and your RTO is four hours, it must be serving customers again by 14:00.
- RPO is about the clock before the failure. If your last good copy is from 09:30 and your RPO is one hour, you are within target. If it is from yesterday, you are not.
A worked example for a 25-person company:
| Workload | If it is down or loses data | Sensible RTO | Sensible RPO |
|---|---|---|---|
| Online ordering | Revenue stops every minute | Under 1 hour | Minutes |
| Internal wiki | Annoying, work continues | 24 hours | 24 hours |
| Accounting system | Painful at month end | 4 to 8 hours | 1 hour |
| Archived contracts | Rarely touched | 48 hours | 1 day |
Those figures are illustrations to show the method, not benchmarks. Your own numbers come from a business impact analysis: for each workload, estimate the hourly cost of downtime and of lost data, then decide what you are willing to pay to reduce them. Google Cloud’s guidance implies this should be done workload by workload rather than as one company-wide target, and warns that smaller targets generally cost more to achieve.
If you want a template for tiering workloads into recovery classes, this walkthrough of RPO and RTO tiering for continuity planners is a practical companion.
The four cloud DR strategies
AWS groups cloud recovery into four approaches that rise in cost and complexity while recovery time and data loss generally fall. Other vendors use different names (cold, warm and hot), but the ladder is the same.

Backup and restore
You copy data and configuration to cloud storage, ideally in a different region, and rebuild infrastructure only when something breaks. It is the cheapest and slowest option. AWS characterizes it as typically giving RPOs measured in hours and RTOs of 24 hours or less, and notes that automated point-in-time backups can bring RPO down to about five minutes in some cases.
Best for: internal tools, archives, development environments and anything that can tolerate a day of downtime. This is where most small businesses should start.
Pilot light
The core pieces, usually the database replica and the deployment definitions, run continuously in the recovery region, while application servers stay switched off. When disaster hits, you start the rest. Recovery takes minutes to hours instead of a day, and the standing cost is small because only the data layer is running.
Warm standby
A scaled-down but fully functional copy of the system runs all the time. On failover you scale it up and redirect traffic. It costs noticeably more than pilot light, but recovery is faster and you can test it continuously because it is already serving a trickle of traffic or synthetic checks.
Multi-site active-active
The workload runs in two or more regions at once and serves live traffic from all of them. Microsoft’s Azure guidance associates mission-critical workloads with RTOs in seconds and RPOs approaching zero, generally requiring active-active multi-region deployment. The price is cost and complexity: data synchronization, conflict handling and double the operational surface.
| Strategy | Typical recovery speed | Relative cost | Complexity | Good fit |
|---|---|---|---|---|
| Backup and restore | Hours (up to 24 hours or less) | Lowest | Low | Internal and archival systems |
| Pilot light | Tens of minutes to hours | Low to moderate | Moderate | Important systems with some tolerance |
| Warm standby | Minutes | Moderate to high | Moderate to high | Customer-facing apps |
| Active-active | Seconds to near zero | Highest | High | Revenue-critical, always-on services |
Microsoft also says business-critical workloads commonly need RTO and RPO measured in minutes, balanced between resilience and cost with warm-standby or active-active designs. Google Cloud adds a useful contrast: infrequently modified compliance data may suit a cold pattern, while customer-facing online services may need a hot one.
In short: you rarely pick one strategy for everything. A sensible mid-size setup uses backup and restore for most systems, warm standby for the two or three that earn money, and nothing fancier.
For a closer look at how these AWS patterns apply to a specific kind of server workload, see this per-asset restore checklist for AI servers.
Backups are not a DR plan
Backups protect data. DR restores a working business. The gap shows up in a few predictable places:
- Backup frequency sets your RPO. Nightly backups mean up to a day of lost work. If that is too much, you need more frequent snapshots or continuous replication. This overview of data backup methods helps compare how often different approaches capture recoverable points.
- Restores are what you test, not backups. A backup you have never restored is a hope.
- Configuration is data too. Infrastructure definitions, DNS records, secrets, identity settings and integration credentials all need to be recoverable. Teams that keep infrastructure as code recover faster because the environment is rebuilt from a file, not from memory.
- Keep a copy the attacker cannot reach. Use a separate account or immutable storage so ransomware or a compromised admin cannot delete backups along with production.
What a cloud disaster recovery plan should contain
AWS advises that a DR plan should be part of the business continuity plan rather than a standalone document, because non-IT dependencies, such as staff access, phone lines or a payment processor, can still stop operations when the servers are fine. It also says a business impact analysis should identify workload impacts and set recovery objectives while weighing disruption probability against recovery cost.
A workable plan fits in a short document with these parts:
- Scope and inventory. Every system, who owns it, where its data lives and what it depends on.
- Business impact analysis. Downtime and data-loss cost per workload.
- Recovery objectives. RTO and RPO per workload, signed off by the business owner, not only IT.
- Chosen strategy per workload. Which rung of the ladder, and why.
- Runbooks. Step-by-step recovery instructions someone unfamiliar with the system could follow at 3 a.m., including declaring a disaster and who has authority to do it.
- Roles and contacts. Primary and backup responders, vendors, and who tells customers and staff.
- Communication templates. A status message you can send in minutes.
- Test schedule and results log. Dates, what was restored, how long it took, what failed.
How to build it in six steps
- List systems and rank them by business impact.
- Set RTO and RPO for each, starting with the top five.
- Match each to a strategy and estimate monthly standby cost plus the cost of a three-day recovery.
- Automate what you can: backup schedules, infrastructure definitions, failover scripts.
- Write runbooks and store them somewhere that does not depend on the systems being recovered.
- Run a restore test, time it, compare to your RTO, fix the gaps, and repeat on a schedule.
Start with the restore test on your most important system. The first one almost always finds a missing credential, an undocumented dependency or a backup that did not capture what you thought it did.
Security, compliance and healthcare data
When you move recovery into someone else’s infrastructure, evaluate the provider’s capabilities, service-level agreements, connectivity, and security and compliance controls. TechTarget notes that cloud services can support regulated uses such as HIPAA and PCI DSS, while some sensitive data may still call for local or owned off-site recovery.
For healthcare organizations in the US, a recovery plan is not optional. The HIPAA Security Rule at 45 CFR 164.308(a)(7) requires covered entities and business associates to maintain a contingency plan, which includes a disaster recovery plan, a data backup plan and an emergency mode operation plan, with periodic testing and a criticality analysis of applications and data as addressable items. In practice that means:
- Anywhere ePHI is stored, including backups and replicas, needs the same safeguards as production: encryption, access controls and audit logging.
- Your cloud provider must sign a business associate agreement before it handles ePHI.
- Recovery tests should be documented, because the documentation is what you show an auditor.
Talk to your compliance officer or healthcare counsel about your specific obligations before finalizing the design. The rule sets what must exist; it does not choose your architecture for you.
Choosing an approach on a small-business budget
A few rules keep the spending sensible:
- Tier your systems. Putting everything on warm standby is the most common way to overspend. Reserve it for what earns revenue.
- Price the bad day, not just the quiet month. As the research above shows, runtime costs during an active recovery can exceed alternatives, so know your worst-case bill.
- Consider hybrid. If one legacy system cannot run in the cloud, keep it on a local replica and put everything else in the cloud.
- Automate before you scale up. A scripted backup-and-restore setup beats a half-managed warm standby.
- Review yearly. Your systems change, and so does the right strategy. Our guide to digital transformation for smaller organizations covers how to sequence this kind of work without overreaching.
If you are deciding whether to build recovery features into custom software or buy a ready-made tool, our piece on custom software costs and when to build versus buy lays out the trade-offs.
Our perspective
Most recovery failures we see are not technology failures. They are planning failures: nobody agreed on what “back online” means, nobody owned the runbook and nobody had ever restored anything.
That is why we start every engagement with deep discovery. Before choosing a cloud pattern we map the workflows the business depends on, the data each one touches and the real cost of an hour offline. Recovery targets set that way are defensible, and the architecture that follows is usually simpler and cheaper than what a vendor would propose by default.
Our engineers bring big-tech experience to this work, and we build to HIPAA and SOC 2 expectations for clients in healthcare and finance. We would rather give a small team a tested backup-and-restore setup with clear runbooks than sell it an active-active design it cannot operate. As the business grows, the design scales with it.
The best DR plan is the one your team has actually rehearsed.
Cameron
Work with CoreWorx on cloud recovery
If you want a second set of eyes on your recovery targets, our Cloud & DevOps and system integration work covers architecture, automation and the runbooks that make recovery repeatable. For teams that want to start small, the Discovery Audit is a $999 one-off engagement that maps your systems and priorities before any build decision. See the Discovery Audit and build options or look at our Cloud & DevOps services.

For healthcare teams, we also build HIPAA-compliant systems, such as the patient portal described in this case study.
FAQ
What is the difference between disaster recovery and backup?
Backup is a copy of data. Disaster recovery is the full capability to restore systems, configuration and access within defined time limits. You need backups for DR, but DR also needs runbooks, a recovery environment, assigned roles and tested procedures.
What are RTO and RPO in simple terms?
RTO is how long you can be offline before the damage becomes unacceptable. RPO is how much recent data you can afford to lose, measured as time since the last recoverable copy. Together they decide how much you should spend on recovery.
Which cloud DR strategy is best for a small business?
Usually backup and restore for most systems, with pilot light or warm standby for the few that bring in revenue. The four options run from backup and restore to active-active, and cost and complexity climb with each step. Pick the lowest tier that meets your RTO and RPO.
How often should we test our disaster recovery plan?
On a schedule you set and keep, with a full restore of your most critical system at least once a year and after any major architecture change. Microsoft recommends scheduled drills, and HIPAA lists periodic testing as an addressable item for covered entities. Record each result and fix what failed.
Sources
- Disaster recovery planning guide, Google Cloud
- Business Continuity Plan (BCP), AWS whitepaper
- Disaster recovery is different in the cloud, AWS whitepaper
- REL13-BP02 Use defined recovery strategies, AWS Well-Architected
- Develop a disaster recovery plan for multi-region deployments, Microsoft Azure
- What is Disaster Recovery? Microsoft Azure
- Cloud, co-location, on-premises and hybrid disaster recovery, PubMed
- Cloud disaster recovery, TechTarget
- HIPAA disaster recovery plan requirements
- RPO and RTO tiering templates
- Per-asset AI server DR checklist
- Examples of data backup methods