Estimation Cheat Sheet and Trade-off Vocabulary
Fluent system design candidates estimate quickly from a few memorized numbers and name trade-offs on the spot. Here are the numbers, the latency table, a six-step estimation recipe with a calculator, and a deck of the trade-offs interviewers expect you to weigh.
🎯 What you'll learn
- Capacity and size numbers to memorize
- Latency numbers and how to use them
- A six-step estimation recipe
- The core trade-offs, practised as flashcards
📚 Before you start
Two things separate fluent system design candidates from the rest: they can estimate quickly with a few memorized numbers, and they have a vocabulary of trade-offs they can name and weigh on the spot. This page collects both.
1Numbers to Know
| Item | Value |
|---|---|
| Seconds per day | ~86,400 ≈ 10⁵ |
| Seconds per month | ~2.6 million ≈ 2.5 × 10⁶ |
| 1 million requests/day | ≈ 12 QPS average |
| Peak vs average | 2–3x typical; 10x for spiky events |
| Char (ASCII) / UUID / timestamp | 1 B / 16 B / 8 B |
| Typical small DB row | 100 B – 1 KB |
| Image (compressed) / short video (1 min, 720p) | 200 KB – 2 MB / ~20–50 MB |
| Single Postgres/MySQL box | ~5–10k simple writes/s, tens of thousands of reads/s (with indexes, good hardware) |
| Redis node | ~100k+ ops/s |
| Kafka partition | ~10 MB/s+ writes |
2Latency Numbers
| Operation | Time |
|---|---|
| L1 cache / main memory reference | ~1 ns / ~100 ns |
| Read 1 MB sequentially from memory | ~10 µs |
| SSD random read | ~100 µs |
| Read 1 MB sequentially from SSD | ~200 µs – 1 ms |
| Round trip within a data center | ~0.5 ms |
| HDD seek | ~5–10 ms |
| Round trip US east ↔ west / transatlantic | ~60–80 ms / ~80–150 ms |
Use these to reason: "a cross-region synchronous write adds ~100 ms, so we'll replicate asynchronously".
3The Estimation Recipe
- Daily active users × actions per user = actions per day.
- ÷ 10⁵ = average QPS; × 2–3 = peak QPS.
- Split reads vs writes (often 10:1 to 100:1 for consumer apps).
- Writes per day × size × retention = storage (add replication ×3).
- QPS × response size = bandwidth.
- Compare with capacities: one DB? a cache? how many servers?
4The Trade-off Vocabulary
Practise naming and explaining each trade-off in about 30 seconds: what you gain, what you pay, and when you'd choose each side.
⚠️ Common Pitfalls
- False precision: 11.574 QPS instead of "about 12".
- Forgetting peak and replication.
- Not using the numbers: estimates must lead to decisions ("so one primary with read replicas is enough").
- Name-dropping trade-offs without saying which side you choose and why.
📋 Quick Reference
1 day ≈ 10⁵ s · 1M req/day ≈ 12 QPS · peak = 2–3x · storage × 3 for replication
Memory 100 ns · SSD read 100 µs · DC round trip 0.5 ms · cross-continent 100 ms
Recipe: DAU × actions → QPS → read/write split → storage → bandwidth → capacity
Trade-off answer: gain · cost · when to choose each side · your choice and why🧠 Key takeaways
- Memorize a few anchors: 1 day ≈ 10⁵ s, 1M req/day ≈ 12 QPS, peak 2–3x.
- Know the latency ladder from memory (100 ns) to cross-continent (100 ms).
- Estimation is useful only when it drives a decision.
- For any trade-off: gain, cost, when to choose each side, and your choice.
🛠️ Practice exercises
Each exercise is a file in this topic folder. Code exercises are self-checking: run them and make the checks pass.
✅ Check yourself
Answer the questions — feedback is instant.