Quick Reference
System Design Cheat Sheet
Every number, formula, and decision tree you need in one place. Print it, bookmark it, review it before your interview.
Latency Numbers Every Engineer Should Know
Jeff Dean's classic numbers, updated for modern hardware (2026).
| Operation | Latency | Category |
|---|---|---|
| L1 cache reference | 0.5 ns | CPU |
| Branch mispredict | 5 ns | CPU |
| L2 cache reference | 7 ns | CPU |
| Mutex lock/unlock | 25 ns | CPU |
| Main memory reference | 100 ns | Memory |
| Compress 1 KB with Snappy | 3 μs | Memory |
| Send 1 KB over 1 Gbps network | 10 μs | Network |
| Read 4 KB randomly from SSD | 150 μs | Disk |
| Read 1 MB sequentially from memory | 250 μs | Memory |
| Round trip within same datacenter | 500 μs | Network |
| Read 1 MB sequentially from SSD | 1 ms | Disk |
| HDD seek | 10 ms | Disk |
| Read 1 MB sequentially from HDD | 20 ms | Disk |
| Send packet CA → NL → CA | 150 ms | Network |
💡 Key insight: Memory is ~100× faster than SSD. SSD is ~100× faster than HDD. Network round-trip in a datacenter ≈ 500μs.
Powers of Two
210
1 Thousand
≈ 1 KB
220
1 Million
≈ 1 MB
230
1 Billion
≈ 1 GB
240
1 Trillion
≈ 1 TB
250
1 Quadrillion
≈ 1 PB
QPS Rules of Thumb
1 web server
1K–10K RPS
1 database (SQL)
5K–30K QPS
1 Redis instance
100K–500K QPS
1 Kafka broker
200K–1M msg/s
1 CDN PoP
100K+ RPS
Availability Nines
| Level | Availability | Downtime / Year |
|---|---|---|
| Two nines | 99% | 3.65 days/year |
| Three nines | 99.9% | 8.77 hours/year |
| Four nines | 99.99% | 52.6 min/year |
| Five nines | 99.999% | 5.26 min/year |
| Six nines | 99.9999% | 31.5 sec/year |
💡 Most SaaS companies target 99.9%–99.99%. Serial components multiply availability (0.999 × 0.999 = 0.998). Parallel (redundant) components: 1 − (1−A)² = higher.
Common Storage Estimates
Back-of-Envelope Formulas
QPS
Daily Active Users × Avg Requests/User / 86,400
Example: 10M × 5 / 86400 ≈ 580 QPS
Peak QPS
QPS × 2–3 (peak factor)
Example: 580 × 3 ≈ 1,740 QPS
Storage (5 yr)
Writes/day × size × 365 × 5
Example: 50M × 250B × 1825 ≈ 22.8 TB
Bandwidth
QPS × avg response size
Example: 1K × 500KB = 500 MB/s
Cache Memory
QPS × response size × cache TTL
Example: 1K × 1KB × 300s = 300 MB
Number of Servers
Peak QPS / single server QPS
Example: 10K / 2K = 5 servers
Quick Decision Trees
45-Minute Interview Checklist
1. Clarify Requirements
Functional & non-functional. Users, scale, latency, consistency.
2. Back-of-Envelope
QPS, storage, bandwidth, memory. Show the math.
3. API Design
Key endpoints, request/response schema, auth.
4. Data Model
Entities, relationships, database choice with justification.
5. High-Level Design
Draw boxes: clients → LB → services → cache → DB → queues.
6. Deep Dives
Pick 2-3 interesting areas. Discuss trade-offs in depth.
7. Bottlenecks & Scaling
Identify bottlenecks. Propose solutions. Discuss limits.
8. Wrap Up
Summarize trade-offs. Mention what you'd improve with more time.
⏱️ Time allocation: Requirements (5 min) → Estimation (5 min) → API/Data Model (5 min) → High-Level Design (10 min) → Deep Dives (15 min) → Wrap Up (5 min)
Common Acronyms
Ready to go deeper?
Explore structured topics with interactive diagrams, quizzes, and code examples.