Planning for the peak
Capacity planning is the practice of estimating how much resource a system needs and provisioning enough to meet demand reliably. The key mistake is planning for the average load when failures and growth happen at the peak.
Why headroom matters
Headroom is the spare capacity you keep beyond normal usage. Running a system at one hundred percent utilization is dangerous: a small traffic spike, a slow garbage collection pause, or a failed node leaves nothing in reserve and latency explodes.
- Keep utilization well below the limit, often around half, so spikes have room.
- Reserve enough that you can lose a node or a region and still serve traffic.
Estimating need
- Measure current load and its growth rate.
- Project to a future horizon and a realistic peak to average ratio.
- Add headroom for failures and the unexpected.
Headroom is not waste; it is the buffer that keeps latency stable and absorbs the surprises that always come.
Key idea
Capacity planning sizes for the peak and keeps deliberate headroom so spikes and node failures do not collapse the system.