Stop 13 · Site Reliability Engineering PracticesTopics · 5/8

Capacity Planning

In short

Running out of capacity during your biggest sale is the worst time to find out. Capacity planning uses today's numbers to predict next month's needs, so you buy the extra servers before the crowd arrives.

Forecasting future resource needs based on growth trends, so infrastructure scales ahead of demand rather than reactively during a crisis — combines monitoring data with basic forecasting.

The mascot stacks extra boxes of sweets on shop shelves under festive lights and garlands.

Think of it like this

The mithai shop before festival week

A sweet shop owner doesn't wait for the festival rush to order more khoya. He looks at last year's sales, adds this year's growth, and stocks up a week early with some extra for surprises. Capacity planning is that same habit for CPUs, memory and database size.

Watch it happen

Will the database disk last?

210

240

275

Apr

May

Jun

1/4The database disk is 500 GB. Here's how full it was at the end of each month so far.

The big ideas

What you need to know

Measure the trend

Use real usage over weeks or months: CPU, memory, requests per second, disk. Guesses are not a plan.

Keep headroom

Plan to act around 60 to 70% full, not at 95%. Spikes and slow purchase approvals eat the gap fast.

Know your events

Festive sales, exam results day or a marketing launch can multiply traffic. Add them on top of normal growth.

Find the real limit

Load test to learn what one server can handle. Then capacity is just expected load divided by that number, plus spare.

See it in code

Walk through the code

excel

1A: Month   B: Peak RPS2A2: Jan    B2: 8203A3: Feb    B3: 9054A4: Mar    B4: 10105 6Growth per month (avg):   =AVERAGE(B3/B2, B4/B3) - 17Forecast for Jun:         =B4 * (1 + B6)^38One server handles:       300 RPS   (from a load test)9Servers needed in Jun:    =ROUNDUP(B7 / 300 / 0.7, 0)

1/4Three months of real peak requests per second from your monitoring.

Assume B6 holds the growth formula and B7 the forecast. Dividing by 0.7 keeps each server at about 70% busy at peak.

Save yourself the pain

Common mistakes

Oops

Relying only on autoscaling and never planning.

Do this instead

Autoscaling has limits: account quotas, database size, warm-up time. Plan the parts that can't scale in seconds.

Oops

Planning from average traffic.

Do this instead

Plan from peaks. Your servers fall over at the busiest five minutes, not on the daily average.

Your mission

Forecast in a spreadsheet

~25 min

Quick check

Disk is at 62% and grows 4% a month. When should you start the upgrade?

Go deeper

1Sketch a simple capacity forecast: current usage trend projected 3-6 months forward.
2Learn to build in buffer capacity for unexpected spikes, not just the forecasted trend line.
3Understand why capacity planning matters even more for systems that can't auto-scale easily.