The 4-Step System Design Approach: From Problem to Solution

The Framework

Whether in an interview or a real greenfield design, a structured approach prevents you from jumping straight to solutions before you understand the problem. The 4-step framework ensures you cover requirements, scale, design, and trade-offs systematically.

Flow:

1. Step 1: Clarify Requirements — Functional + non-functional requirements; out-of-scope boundaries
2. Step 2: Estimate Scale — DAU, QPS, storage, bandwidth — back-of-envelope calculations
3. Step 3: High-Level Design — Core components, data flow, APIs, database schema
4. Step 4: Deep Dive — Zoom in on hard/critical components; discuss trade-offs

Step 1 — Clarify Requirements

• Functional: What must the system do? (user stories, core features)
• Non-functional: Scale, availability (SLO?), latency (p99?), consistency, security, compliance
• Constraints: Must use existing systems? Specific cloud provider? Budget?
• Out of scope: Explicitly state what you are NOT designing

Step 2 — Estimate Scale

// Example: Design Twitter (1B users, 300M DAU)

// Reads vs Writes (read-heavy: 100:1)
Write QPS:  300M DAU * 1 tweet/day / 86,400s ≈ 3,500 QPS
Read QPS:   3,500 * 100                       ≈ 350,000 QPS

// Storage (tweets retained forever)
Tweet size: 280 chars * 2 bytes + metadata    ≈ 1 KB
Daily writes: 3,500 QPS * 86,400s            = 300M tweets/day
Daily storage: 300M * 1 KB                   = 300 GB/day
5-year storage: 300 GB * 365 * 5             ≈ 550 TB

// Bandwidth
Read bandwidth: 350,000 QPS * 1 KB           = 350 MB/s  ← need CDN
Write bandwidth: 3,500 QPS * 1 KB            = 3.5 MB/s  ← manageable

Step 3 — High-Level Design

• Draw the main components: clients, load balancers, services, caches, databases, queues
• Define the primary APIs (endpoint, method, request/response)
• Choose database(s) and justify: SQL vs NoSQL, what schema
• Identify the core data flow for the most critical use case (e.g. post tweet, see feed)

Step 4 — Deep Dive

• Pick the hardest 2–3 sub-problems and solve them in depth
• Common deep-dives: feed generation, search, notifications, chat, storage, consistency
• Discuss trade-offs explicitly: "I chose X over Y because… the downside is…"
• Address bottlenecks identified in estimation: if 350K QPS read, explain how cache + CDN handles it

TIP: In interviews: spend 5 min on Step 1, 5 min on Step 2, 15 min on Step 3, and 15 min on Step 4. Interviewers want to see your thought process and trade-off reasoning more than a perfect answer.