# Peaks, Spikes and Seasonality — Cost & Capacity Estimation

Source: https://www.skillbyai.com/en/capacity-estimation/ce-peak-traffic

> Convert average QPS into a design peak using daily curves, event spikes and a safety margin.

## Design for the peak, not the average

Traffic is never flat. Usage follows a **daily curve** with quiet nights and busy evenings, a **weekly pattern**, and **event spikes**: a cricket final, a festive sale, exam results, a push notification sent to every user at once. Systems fail at peaks, so capacity is planned for **peak QPS = average QPS x peak factor**. For organic daily traffic, a peak factor of roughly 2 to 3 is a common starting assumption; for flash events it can be far higher, and the only reliable source is your own historical metrics. Short spikes also differ from sustained peaks: a burst lasting seconds can be absorbed by queues and autoscaling headroom, while an evening peak lasting hours must be served by provisioned capacity. When you lack data, state a peak factor explicitly and explain why you chose it.

## Peak from an hourly profile

If you know what share of daily traffic lands in the busiest hour, you can derive the peak factor instead of guessing.

```python
daily_requests = 300_000_000
busiest_hour_share = 0.12      # 12% of the day's traffic in one hour

avg_qps = daily_requests / 86_400
peak_qps = daily_requests * busiest_hour_share / 3_600
print(f"average QPS : {avg_qps:,.0f}")
print(f"peak QPS    : {peak_qps:,.0f}")
print(f"peak factor : {peak_qps / avg_qps:.1f}x")
```

Output:

```
average QPS : 3,472
peak QPS    : 10,000
peak factor : 2.9x
```

## Metro at rush hour

A metro line designed for the average number of passengers per hour would be unusable at 9 a.m. Trains, platforms and gates are sized for rush hour, and extra trains are added for big events. Servers are planned the same way.

## Notifications create self-made spikes

Sending a push notification to all users at the same second can create a spike far above your normal peak. Staggering sends over several minutes is a cheap capacity fix.

**Quiz:** If 10% of a day's requests arrive in the busiest hour, what is the peak-to-average factor?

- [x] 2.4x
- [ ] 10x
- [ ] 1.1x
- [ ] 24x

*Answer:* 2.4x. Spread evenly, each hour would carry 1/24 (about 4.2%) of traffic. 10% / 4.17% = 2.4x.
