Skip to content

Excalidraw runs in the browser; or install draw.io desktop

Architect, this is the proving ground. You have mastered patterns, domains, services, events, gateways, and scale - and now you must wield them all at once, on a whiteboard, under a clock, while a stranger watches and asks "what happens when this fails?" The System Design Interview is the capstone of the Citadel: not a test of trivia, but of whether you can take a vague prompt and reason your way to a credible, defensible architecture out loud.

Whether you are preparing for a senior or staff interview, or you simply want a repeatable way to approach any "design X" problem, this quest forges the one thing that separates those who freeze from those who flow: a framework. With it, "Design a URL shortener" stops being a panic and becomes a checklist.

📖 The Legend Behind This Quest

The open-ended design question terrifies because it has no single right answer. But interviewers are not looking for the answer - they are watching your process. Do you clarify the problem before diving in? Do you estimate scale before choosing technology? Do you name trade-offs without being prompted? Do you find your own design's bottlenecks?

A framework turns chaos into a script you can run on any prompt: clarify requirements, estimate the load, sketch the high-level design, deep-dive a component, then identify and address bottlenecks. The estimation - the famous "back-of-the-envelope" math - tells you whether you need one server or a thousand. Master the framework and the math, and the interview becomes a conversation you lead.

🎯 Quest Objectives

By the time you complete this epic journey, you will have mastered:

Primary Objectives (Required for Quest Completion)

  • A Repeatable Framework - Run any design prompt through the same five-step structure
  • Requirements Gathering - Separate functional from non-functional requirements and scope down
  • Back-of-the-Envelope Estimation - Compute QPS, storage, and bandwidth in your head
  • Trade-off Articulation - Name the cost of every choice before you are asked

Secondary Objectives (Bonus Achievements)

  • Two Worked Designs - Design a URL shortener and a news feed end to end
  • Bottleneck Hunting - Find and address your own design's weak point
  • Communication Under Pressure - Think aloud, draw clearly, and stay calm

Mastery Indicators

You'll know you've truly mastered this quest when you can: - [ ] Run a full design from prompt to bottlenecks in 45 minutes - [ ] Estimate QPS and storage to the right order of magnitude - [ ] Volunteer a trade-off ("I'd choose AP here because...") unprompted - [ ] Recover gracefully when the interviewer adds a new constraint

🗺️ Quest Prerequisites

📋 Knowledge Requirements

🛠️ System Requirements

  • Modern operating system (Windows 10+, macOS 10.14+, or Linux)
  • A whiteboard or a diagramming tool (Excalidraw, draw.io)
  • A timer to practice under interview conditions

🧠 Skill Level Indicators

This ⚔️ Epic quest expects: - [ ] You have built or operated a real multi-component system - [ ] You can estimate orders of magnitude quickly - [ ] Ready for 5-6 hours of focused practice

🌍 Choose Your Adventure Platform

This quest is about thinking and communicating, so the "platform" is your diagramming surface. Pick whatever lets you sketch boxes and arrows fast.

🍎 macOS Kingdom Path

Click to expand macOS instructions 
# Excalidraw runs in the browser; or install draw.io desktop
brew install --cask drawio
# A simple QPS/storage estimator you can extend (see Chapter 2)
python3 -c "print('Ready to estimate at scale')"

🪟 Windows Empire Path

Click to expand Windows instructions 
winget install JGraph.Draw
python -c "print('Ready to estimate at scale')"

🐧 Linux Territory Path

Click to expand Linux instructions 
sudo apt update && sudo apt install -y python3
# Or use Excalidraw in any browser — no install needed
python3 -c "print('Ready to estimate at scale')"

☁️ Cloud Realms Path

Click to expand Cloud/Container instructions 
# Excalidraw.com and draw.io are fully browser-based — ideal for a remote
# interview where you share a virtual whiteboard.
echo "Open https://excalidraw.com in your browser"

🧙‍♂️ Chapter 1: The Framework - A Script for Any Prompt

The single most valuable interview asset is a structure you run every time, so you never stare at a blank board.

⚔️ Skills You'll Forge in This Chapter

  • The five-step design framework
  • Clarifying scope before designing
  • Functional vs. non-functional requirements

🏗️ The Five Steps

graph TB
    R[1. Requirements<br/>clarify + scope] --> E[2. Estimation<br/>QPS, storage, bandwidth]
    E --> H[3. High-Level Design<br/>boxes and arrows]
    H --> D[4. Deep Dive<br/>one component in detail]
    D --> B[5. Bottlenecks<br/>find + address]
    B -.iterate.-> H
    style R fill:#2196F3,color:#fff
    style E fill:#FF9800,color:#fff
    style H fill:#4CAF50,color:#fff
    style D fill:#7c3aed,color:#fff
    style B fill:#ef4444,color:#fff
  1. Requirements - Ask questions. Who uses this? How many? What must it do (functional) and how well - latency, availability, consistency (non-functional)? Scope ruthlessly; a 45-minute interview cannot design everything.
  2. Estimation - Compute the scale so your design is grounded, not hand-wavy (Chapter 2).
  3. High-Level Design - Draw clients, a gateway/load balancer, services, databases, and caches. Keep it simple first.
  4. Deep Dive - The interviewer picks (or you offer) one component to detail: the data model, the sharding scheme, the cache strategy.
  5. Bottlenecks - Critique your own design. Where does it fall over at 10x? What is the single point of failure? This step impresses most.

🏗️ Functional vs. Non-Functional

Functional ("what it does"):
  - Shorten a long URL to a short code
  - Redirect a short code to the original
Non-functional ("how well"):
  - Read-heavy (100:1 reads to writes)
  - < 100ms redirect latency
  - High availability (a dead shortener breaks every link)
  - Short codes must not collide

🔍 Knowledge Check: The Framework

  • Why clarify requirements before drawing anything?
  • What is the difference between a functional and a non-functional requirement?
  • Why is the bottleneck step the one that most impresses interviewers?

🧙‍♂️ Chapter 2: Estimation - Back-of-the-Envelope Math

Numbers turn a vague design into a grounded one. You do not need precision - you need the right order of magnitude, computed in seconds.

⚔️ Skills You'll Forge in This Chapter

  • Powers-of-two and powers-of-ten cheat sheet
  • QPS, storage, and bandwidth estimates
  • Letting numbers drive design choices

🏗️ The Cheat Sheet

Time:    1 day ≈ 86,400 s ≈ 10^5 s     |  1 month ≈ 2.5 × 10^6 s
Data:    1 KB = 10^3 B, 1 MB = 10^6 B, 1 GB = 10^9 B, 1 TB = 10^12 B
Latency: memory ~100 ns | SSD ~100 µs | network round trip (same DC) ~0.5 ms
         disk seek ~10 ms | cross-continent round trip ~150 ms

🏗️ Worked Example: A URL Shortener

# Estimation for a URL shortener — the kind of math you do out loud.
WRITES_PER_MONTH = 100_000_000          # 100M new URLs/month (given/assumed)
SECONDS_PER_MONTH = 2_500_000           # ~2.5M s/month

write_qps = WRITES_PER_MONTH / SECONDS_PER_MONTH      # ≈ 40 writes/sec
read_qps  = write_qps * 100                           # 100:1 read ratio → ≈ 4,000 reads/sec

BYTES_PER_URL = 500                      # short code + long URL + metadata
storage_per_month = WRITES_PER_MONTH * BYTES_PER_URL  # = 5 × 10^10 B = 50 GB/month
storage_5yr = storage_per_month * 12 * 5              # = 3 TB over 5 years

print(f"Writes: ~{write_qps:.0f}/s | Reads: ~{read_qps:.0f}/s")
print(f"Storage: ~{storage_per_month/1e9:.0f} GB/month, ~{storage_5yr/1e12:.0f} TB / 5yr")

These numbers drive the design: 4,000 read QPS and a 100:1 ratio scream "cache aggressively." 3 TB over five years fits comfortably on a single sharded database - you do not need exotic storage. Stating this reasoning aloud is exactly what interviewers reward.

🔍 Knowledge Check: Estimation

  • Roughly how many seconds are in a day? A month?
  • Given 100:1 reads-to-writes, what does that imply for your design?
  • Why do interviewers care about estimates more than exact numbers?

🧙‍♂️ Chapter 3: A Full Walkthrough and Articulating Trade-offs

Tie it together with one complete design, narrating trade-offs the whole way - the skill that distinguishes a senior candidate.

⚔️ Skills You'll Forge in This Chapter

  • A complete URL-shortener design
  • Naming trade-offs unprompted
  • Handling a curveball constraint

🏗️ Designing the URL Shortener

High-level design (after the estimation above):

graph LR
    Client --> LB[Load Balancer]
    LB --> API[Shortener Service<br/>stateless, N replicas]
    API --> Cache[(Redis cache<br/>hot short→long)]
    API --> DB[(Sharded key-value store<br/>short_code → long_url)]
    Cache -. miss .-> DB
    style Cache fill:#FF9800,color:#fff
    style DB fill:#7c3aed,color:#fff
    style API fill:#4CAF50,color:#fff

Deep dive - generating short codes. Two options, and naming the trade-off is the point:

import hashlib, string

ALPHABET = string.ascii_letters + string.digits   # 62 chars → 62^7 ≈ 3.5 trillion codes

# Option A — hash the URL, take 7 chars. Risk: collisions need handling.
def code_by_hash(url: str) -> str:
    digest = hashlib.sha256(url.encode()).hexdigest()
    n = int(digest, 16)
    out = []
    for _ in range(7):
        n, rem = divmod(n, 62)
        out.append(ALPHABET[rem])
    return "".join(out)

# Option B — a global counter, base62-encoded. No collisions, but needs a
# distributed id generator (e.g. a range-allocator or Snowflake-style ids).
def code_by_counter(counter_id: int) -> str:
    if counter_id == 0:
        return ALPHABET[0]
    out = []
    while counter_id:
        counter_id, rem = divmod(counter_id, 62)
        out.append(ALPHABET[rem])
    return "".join(reversed(out))

"I'd lean toward the counter approach: it guarantees uniqueness without collision-retry logic, at the cost of needing a distributed ID generator. The hash approach is simpler but I'd have to handle collisions, which complicates the write path." — that sentence is what earns the offer.

Bottlenecks. The database write path and the single-region cache. Mitigations: shard by short code, add read replicas, and use a CDN/edge cache for the hottest links.

🏗️ Articulating Trade-offs

For every decision, say the alternative and its cost: SQL vs. NoSQL (consistency vs. scale), sync vs. async (simplicity vs. coupling), CP vs. AP (correctness vs. availability). When the interviewer adds a constraint ("now make it analytics-friendly"), fold it into the framework instead of panicking - add an event stream feeding a data warehouse.

🔍 Knowledge Check: Walkthrough

  • What trade-off separates hash-based from counter-based code generation?
  • Why does a 100:1 read ratio justify a CDN/edge cache for hot links?
  • How should you respond when the interviewer adds a new requirement mid-design?

🎮 Mastery Challenges

🟢 Novice Challenge: Run the Framework

Objective: Take a fresh prompt ("design a pastebin") and write the five steps as bullet points in 15 minutes.

Requirements: - [ ] Functional and non-functional requirements listed - [ ] One estimation (QPS or storage) - [ ] A high-level box-and-arrow sketch

Validation: A peer can follow your reasoning from prompt to sketch.

🟡 Intermediate Challenge: Estimate Under Time

Objective: For "design a news feed," compute write QPS, read QPS, and 5-year storage in under 5 minutes.

Requirements: - [ ] State your assumptions (users, posts/day, fanout) - [ ] Reach the right order of magnitude - [ ] Name one design choice the numbers force

Validation: Your estimate is within 10x of a worked reference and the design choice follows from it.

🔴 Advanced Challenge: Full Mock Design

Objective: Run a complete 45-minute mock for a prompt of your choice and write it up.

Requirements: - [ ] All five framework steps, on a clock - [ ] At least three trade-offs articulated unprompted - [ ] Your own bottleneck analysis with mitigations

Validation: The write-up reads like a design a staff engineer would approve.

🏆 Quest Rewards & Achievements

🎖️ Badges Earned: - 🏆 Whiteboard Warlord - You run a design interview with calm, repeatable structure - 🧭 Master of Trade-offs - You name the cost of every choice before being asked

🛠️ Skills Unlocked: - Capacity Estimation - Back-of-the-envelope QPS, storage, and bandwidth - Structured System Design - The five-step framework for any prompt

🔓 Unlocked Quests: - Technical Leadership - Lead the teams that build these systems

📊 Progression Points: +110 XP

🗺️ Next Steps in Your Journey

Continue the Main Story: - 🎯 Technical Leadership - From designing systems to leading those who build them

Explore Side Adventures: - ⚔️ Scaling Strategies - Sharpen the scaling tools you used here - ⚔️ Microservices Architecture - Revisit decomposition with interview eyes

Character Class Recommendations

💻 Software Developer: Continue to Technical Leadership 🏗️ System Engineer: Practice more prompts from the System Design Primer 📊 Data Scientist: Apply the framework to data-pipeline design prompts

📚 Resources

Official Documentation

Community Resources

Learning Materials

🤝 Quest Completion Checklist

  • ✅ Completed all primary objectives
  • ✅ Ran a full mock design under a 45-minute clock
  • ✅ Answered all knowledge check questions
  • ✅ Completed at least one mastery challenge
  • ✅ Explored the resource library
  • ✅ Identified your next quest in the journey

🕸️ Knowledge Graph

Structured wiki-links connect this quest to the IT-Journey knowledge graph. Open the Obsidian Graph View to explore connections.

Level hub: [[Level 1110 - Architecture & Design Patterns]] Overworld: [[🏰 Overworld - Master Quest Map]] Prerequisites: [[Scaling Strategies: Horizontal Growth, Caching, and CAP]] · [[Microservices Architecture: Decomposing the Monolith]] Unlocks: [[Technical Leadership]] Obsidian docs: [[Obsidian Knowledge Graph and Wiki Links]]