Back of the Envelope Calculation

Back of the Envelope Calculation

Before designing anything, estimate the scale you're actually designing for. It's easy
to over-engineer for scale you'll never hit, or under-engineer for scale you will.

The basic recipe

1. Estimate daily active users (DAU) and their usage pattern (requests/user/day).
2. Convert to average QPS: DAU × requests/user/day ÷ 86,400.
3. Estimate peak QPS — usually 2–5× average, depending on how spiky traffic is.
4. Estimate storage: records/day × size/record × retention period.
5. Estimate bandwidth: QPS × average payload size.

Example: a URL shortener

• 100M new short URLs created per month → ~40 writes/sec average.
• 100:1 read:write ratio (people click far more than they create) → ~4,000 reads/sec average.
• Each record ~500 bytes → 100M × 500B ≈ 50 GB/month, or ~600 GB/year before any
  cleanup — cheap to store, but tells you an index needs to fit comfortably in memory
  for low-latency redirects.

Numbers worth memorizing

• L1 cache reference: ~1 ns · Main memory reference: ~100 ns
• Round trip within same datacenter: ~0.5 ms · Cross-region round trip: ~50–150 ms
• Reading 1 MB sequentially from SSD: ~1 ms · From spinning disk: ~20 ms

These orders of magnitude matter more than the exact figures — they tell you where
your latency budget is going to be spent.