Track usage trends (CPU, memory, disk, bandwidth) and project growth. Plan 20–30% headroom for peaks. Consider auto-scaling or reserved capacity. Review capacity quarterly and before big campaigns.
What to track
- Metrics: CPU, memory, disk, and bandwidth over time. Use monitoring (Prometheus, cloud metrics, etc.) and keep at least 6–12 months of history. Identify growth rate and seasonality.
- Projection: Extrapolate trend (linear or growth curve). Factor in planned features, marketing campaigns, or known events. Add buffer for uncertainty.
- Bottlenecks: Which resource hits limits first? Plan to add capacity there (vertical scale, more instances, or more disk/bandwidth) before it becomes critical.
Headroom and peaks
- 20–30% headroom: Do not run at 100% utilization. Peaks (traffic spike, backup, batch job) need room. Running near capacity increases latency and risk of outage.
- Peak planning: Know your peak periods (e.g. Black Friday, product launch). Plan capacity so you stay within headroom during peak. Consider auto-scaling to absorb spikes.
- Reserved capacity: For predictable growth, reserved instances or committed use can reduce cost. Balance with flexibility (e.g. can you scale down if demand drops?).
Review and actions
- Quarterly: Review usage and projections. Adjust capacity (resize, add nodes, upgrade disk) before you run out. Document decisions and assumptions.
- Before campaigns: Before a big launch or campaign, run load tests and ensure capacity (and headroom) are sufficient. Have a rollback or scale-up runbook.
- Auto-scaling: If your stack supports it, use auto-scaling for variable workloads. Set min/max and scale triggers; monitor to avoid thrashing or cost surprises.
Summary
Track usage trends and project growth. Plan 20–30% headroom for peaks. Consider auto-scaling or reserved capacity. Review capacity quarterly and before big campaigns.
