DR standby environment cost: cold, warm, and hot compared

Quick answer

DR cost scales with recovery speed. Backup and restore costs a few percent of production and recovers in hours. Pilot light keeps data replicated with minimal compute and costs 10 to 20 percent, recovering in 30 to 60 minutes. Warm standby runs a scaled down copy at 30 to 50 percent and recovers in 5 to 15 minutes. Hot standby costs 100 percent or more and recovers in seconds. Pick from a stated recovery objective, not from instinct, since the gap between warm and hot is often the largest avoidable line in an infrastructure budget.

Disaster recovery is the one area where teams reliably overbuy, because nobody wants to be the person who argued for the cheaper option at the post incident review. The result is hot standby environments protecting workloads whose real recovery objective is an hour.

The decision gets much easier when the options are priced next to the recovery time each one buys.

The four strategies

Strategy	Cost vs production	Recovery time	Data loss
Backup and restore	About 2 to 8 percent	2 to 12 hours	Up to backup interval
Pilot light	About 10 to 20 percent	30 to 60 minutes	Seconds to minutes
Warm standby	About 30 to 50 percent	5 to 15 minutes	Seconds
Hot standby, active-active	100 percent or more	Seconds	Near zero

On a 20,000 dollar a month production estate, that is roughly 800 dollars for backup and restore, 3,000 for pilot light, 8,000 for warm standby, and 20,000 or more for hot. The difference between warm and hot is 144,000 dollars a year, which is worth a conversation about whether 15 minutes is acceptable.

Start from a written objective

Recovery time objective and recovery point objective are business decisions, not engineering preferences. Ask the people who own the product how long the service can be unavailable and how much data loss is tolerable, in numbers, and write the answers down.

The answers vary enormously by system. A payments path may genuinely need seconds. An internal reporting tool can be down for a day. Applying one DR posture to everything means either overspending on the reporting tool or underprotecting payments, and usually both at once.

What pilot light actually keeps running

Pilot light is the most underused option. Keep data continuously replicated to the second region, keep the network, identity, and base configuration in place, and keep compute at zero or near zero. On failover, Terraform or an autoscaling group brings the compute up against data that is already there.

Cost is dominated by storage replication and cross region transfer rather than compute, which is why it lands at 10 to 20 percent. The recovery time is set by how fast you can start compute and how well tested the process is, typically 30 to 60 minutes, and the second half of that sentence matters more than the first.

The costs beyond duplicated compute

Cross region replication is the DR expense people forget. Continuous database replication and object storage replication both charge per GB transferred, and for a data heavy service that can exceed the standby compute cost. A database replicating 3 TB a month across regions at 0.02 dollars per GB is about 60 dollars, but at 30 TB it is 600 dollars, plus the storage at the far end.

Duplicate storage is charged in full, since the standby copy is real data. Add the load balancers, NAT gateways, and monitoring in the DR region, which run whether or not traffic arrives. The replication mechanics are covered incross region replication cost.

Make the standby cheaper without losing the objective

Several adjustments cut the standby bill without changing recovery time much. Run standby compute on smaller instance classes and scale up on failover, since the first minutes after failover rarely need full capacity immediately. Use single-AZ in the standby region, because the disaster you are protecting against is a regional one. Shorten standby log retention and skip non critical components like analytics pipelines.

These commonly take a warm standby from 50 percent of production to around 30 percent while keeping the recovery time inside the objective.

Test it, and price it in code

An untested DR environment is an expensive assumption. If you have never failed over, you do not have a recovery time, you have a hope. Run a real failover at least twice a year and measure. Teams that test usually discover their pilot light recovers faster than they feared, which lets them downgrade from warm and keep the saving.

Because the DR environment is Terraform, its cost is estimable before it is built. c3x prices the plan with no cloud credentials, so you can compare a pilot light module against a warm standby module as two concrete monthly numbers and take the comparison to the people who own the recovery objective. That turns a debate about risk appetite into a decision with a price attached, which is the only way it ever gets settled.

FAQ

How much does a DR standby environment cost?

As a share of production: backup and restore about 2 to 8 percent with 2 to 12 hour recovery, pilot light 10 to 20 percent with 30 to 60 minute recovery, warm standby 30 to 50 percent with 5 to 15 minute recovery, and hot active-active 100 percent or more with seconds. On a 20,000 dollar a month production estate the warm to hot gap alone is about 144,000 dollars a year.

What is a pilot light DR strategy?

Data is continuously replicated to the second region and the network, identity, and base configuration stay in place, while compute sits at zero or near zero. On failover, Terraform or an autoscaling group starts compute against data that is already present. Cost is dominated by replication and storage rather than compute, which is why it lands at 10 to 20 percent of production.

What DR costs do teams usually forget?

Cross region replication transfer, which is charged per GB and for data heavy services can exceed the standby compute cost, duplicate storage charged in full because the standby copy is real data, and the always-on supporting infrastructure in the DR region such as load balancers, NAT gateways, and monitoring that run whether or not traffic ever arrives.

How can I make a standby environment cheaper?

Run standby compute on smaller instance classes and scale up on failover, since the first minutes rarely need full capacity. Use single-AZ in the standby region, because the disaster being protected against is regional. Shorten standby log retention and skip non critical components like analytics. These usually take a warm standby from 50 percent of production to about 30 percent.

How do I choose between DR strategies?

Start from a written recovery time and recovery point objective supplied by the people who own the product, then price each option against it. Because the DR environment is Terraform, C3X can estimate a pilot light module and a warm standby module from their plans with no cloud credentials, turning a debate about risk appetite into two concrete monthly numbers.

What to do next

Compare DR postures as two real numbers. C3X prices Terraform plans statically, no cloud credentials. Try the quickstart.