Capacity planning basics

Topic: Monitoring basics

Use historical metrics and growth trends to plan for future capacity: when will disk, CPU, or memory be exhausted? Use this when sizing new systems or when deciding when to scale or upgrade to avoid running out of resources.

Intent: How-to

Quick answer

  • Collect metrics over time (weeks to months). Plot usage (disk, CPU, memory) vs time; fit a trend (linear or growth rate). Estimate when usage will reach a threshold (e.g. 80% disk, sustained high CPU).
  • Include growth events: new features, more users, or seasonal peaks. Plan buffer (e.g. scale when trend hits 70% in 30 days). Use capacity headroom alerts (e.g. 'disk will be full in 14 days at current rate').
  • Document assumptions and review periodically. Automate where possible (e.g. Prometheus recording rules for trend); present to stakeholders so scaling or upgrade is scheduled before crunch.

Prerequisites

Steps

  1. Gather history

    Export or query metrics for disk, CPU, memory over 30–90 days. Identify trend: is usage growing linearly or in steps? Note any one-off events (migrations, traffic spikes).

  2. Project and threshold

    Extrapolate trend (simple linear or average growth rate). Calculate when usage will reach 80% (or your threshold). Add buffer (e.g. act when projection says 60 days to 80%).

  3. Plan action

    Schedule scale-up, upgrade, or cleanup before the projected date. If growth is variable, use a conservative estimate or alert when 'days until full' drops below 30.

  4. Review and automate

    Review projection monthly or quarterly; update assumptions. Use predictive alerts (e.g. 'disk full in 14 days') if your monitoring supports it; otherwise track in a spreadsheet or report.

Summary

Use historical metrics to project when resources will be exhausted; plan scaling or cleanup before that point. Use this for capacity planning and to avoid surprise shortages.

Prerequisites

Steps

Step 1: Gather history

Collect disk, CPU, and memory metrics over a sufficient period; identify trends and one-off events.

Step 2: Project and threshold

Extrapolate to estimate when usage will reach your threshold; add a safety buffer.

Step 3: Plan action

Schedule scaling, upgrade, or cleanup before the projected date.

Step 4: Review and automate

Review projections regularly; use predictive alerts if available.

Verification

  • You have a projection and a planned action date; projections are updated and reviewed.

Troubleshooting

No history — Start collecting; use industry or workload benchmarks as a temporary proxy. Trend not linear — Use step or seasonal model; or use a conservative upper bound.

Next steps

Continue to