Don't Panic — Infrastructure Capacity Planning

Infrastructure capacity planning is the discipline of deciding what a system must have ready before demand arrives. It is not, despite the efforts of countless dashboards, a ritual in which a CPU line is stared at until it confesses. The useful question is whether the service can complete the work people bring it while its latency, errors, availability, and allowed degradation remain acceptable.

The first durable idea is the service target. A request rate on its own is only a large number wearing a name badge. Add the behavior that must hold at that rate, and it becomes a boundary you can test. The second is the demand scenario: normal traffic, a peak, a launch, or the traffic left after some defined loss. History helps with ordinary growth, but it is notoriously poor at mentioning the campaign someone already scheduled.

Next comes the tested limit, which is the most valuable number in the room and also the least portable. Apply representative load, watch throughput, latency, errors, queues, and resource limits, then find the highest load that still meets the target. Run a gradual ramp and a sudden spike. Systems have a special talent for behaving politely while you approach them and dramatically when you arrive all at once.

The arithmetic is pleasantly compact. Divide forecast peak demand by the tested capacity of one unit, then round up. The argument begins immediately afterward. A failure policy says what loss the plan must survive: an instance, a zone, a dependency, or something less conveniently named. A bottleneck is the first constraint that prevents more useful work, and it may be a database connection pool or provider quota rather than the application servers everyone has counted with such affection.

Autoscaling can add supply, but it cannot travel back through its own startup time, quota, or dependency limit. When supply cannot arrive in time, queues, throttles, rate limits, load shedding, and selective degradation become the ways to preserve the work that matters most. None is a decorative emergency lever; each needs defined behavior and a test.

Read the Intro for the full loop from targets to provisioning. Use Slides for the relationships and decisions at a glance. Keep the Cheatsheet nearby when writing the record or checking a plan, and use the Practice tab to turn a scenario into a calculation and an action date. The Quiz is where the apparently harmless server count discovers that it has been asked to explain itself.

Don't Panic — Infrastructure Capacity Planning

Relevant careers