SRE Capacity Planner
A capacity calculator that converts forecast peak demand into a deployment requirement with growth, operational headroom and loss of one availability zone included.
Who this Lab is for
Designed for
- SRE and capacity engineers
- Platform teams sizing stateless services
- Technical leads preparing scaling or launch reviews
Use it when
- Planning a launch or seasonal traffic event
- Reviewing whether a service survives a zone failure
- Turning load-test evidence into a capacity request
A complete run, step by step
Measure peak demand
Use an observed or justified forecast peak rather than a daily average.
Enter safe instance throughput
Use sustained load-test performance below the breaking point.
Apply growth and headroom
Separate expected growth from operational reserve for spikes and noisy neighbours.
Review failure capacity
Confirm the remaining zones can carry demand when one zone is unavailable.
What you will need
Prepare the following information before starting. Use measured evidence where possible; defaults are examples and should not be treated as recommendations.
Peak requests per second
Measured or forecast peak load.
Safe RPS per instance
Use the sustained load-test result, not the breaking point.
Forecast growth
Expected growth over the planning horizon.
Operational headroom
Capacity reserved for spikes and noisy neighbours.
Availability zones
Planner assumes one zone may be unavailable.
What the result tells you
Your report includes
- Forecast peak demand
- Steady-state instance requirement
- Failure-domain-safe capacity recommendation
How it is determined
The model grows current peak demand, applies operational headroom and divides it by safe per-instance throughput. It then increases the fleet so the surviving availability zones can carry the calculated requirement.
The formula is deterministic but assumes uniform instances, even zone distribution and a validated sustained throughput rate.
Model assumptions
- • Per-instance safe throughput is sustained and repeatable.
- • Traffic and replicas distribute evenly across zones.
- • One availability zone is the modelled failure domain.
Authoritative references
Three-zone API fleet
Situation
Peak demand is 12,000 RPS, safe throughput is 650 RPS per instance, forecast growth is 25% and headroom is 30%.
Result
The model sizes for forecast demand plus reserve and adds sufficient instances for the remaining two zones to carry traffic after one-zone loss.
Use the result with engineering judgement
- The model assumes instances have broadly uniform capacity.
- It does not model database, network or downstream bottlenecks.
- Autoscaling speed and startup time require separate validation.
Questions before you begin
Why not use maximum benchmark throughput?
Breaking-point throughput leaves no room for latency variance, dependency slowdown or noisy neighbours. Use a sustained safe rate.
Is headroom the same as growth?
No. Growth changes expected demand; headroom is reserve above that demand for operational uncertainty.
Does this work for serverless services?
The principles still apply, but concurrency limits, cold starts and provider quotas need a service-specific model.
Capacity Planner is under review
This legacy judgement-based Lab has been retired. Existing saved reports remain available, but new execution is disabled.
Open deterministic utilities