System Design Basics: Building Scalable Applications

System design matters for building applications that can handle millions of users. Here are the fundamental concepts.

Core Principles

1. Scalability

Horizontal scaling means adding more servers.

Vertical scaling means giving each server more CPU, RAM, or disk.

2. Load Balancing

Distribute traffic across multiple servers.

• Round robin walks servers in order.
• Least connections prefers the least busy backend.
• IP hashing pins a client IP to a stable backend when you need stickiness.

3. Caching

Store frequently accessed data in fast storage.

# Example: Redis caching
import redis

cache = redis.Redis()

def get_user(user_id):
    # Check cache first
    cached = cache.get(f"user:{user_id}")
    if cached:
        return json.loads(cached)
    
    # Fetch from database
    user = db.get_user(user_id)
    
    # Store in cache
    cache.setex(f"user:{user_id}", 3600, json.dumps(user))
    
    return user

Database Design

SQL vs NoSQL

SQL (relational) fits when you need ACID guarantees, rich joins, and a strict schema you can reason about with the whole team.

NoSQL fits when you need flexible documents, easier horizontal scale-out, or very high write throughput, and you accept different consistency tradeoffs.

Replication

• Master-Slave. Read from replicas
• Master-Master. Write to any node

Key Components

1. CDN for static assets at the edge
2. Message queues for async work (RabbitMQ, Kafka, and similar tools)
3. Microservices when you need to split ownership and deploy independently
4. API gateways as a single front door for clients

Design Process

1. Requirements. Functional and non-functional
2. Capacity estimation for traffic, storage, and bandwidth
3. API design for endpoints and data models
4. Database design for schema and partitioning
5. Component design for the big moving parts
6. Scaling plans for the bottlenecks you expect first

Start with these fundamentals and build from there!

How to practice this skill

Treat this topic as a loop rather than a fact to memorize. First, name the pattern in plain English. Then write the smallest version that proves you understand the invariant. Only after that should you chase speed. That order matters because most interview mistakes are not typing mistakes; they are recognition mistakes. You reach for the right data structure too late, or you optimize before you have named what must stay true.

For system design practice, the useful repetition is spaced. Solve one problem slowly, rewrite the solution from memory the next day, then do a timed variant after the idea feels boring. The second pass is where the pattern moves from recognition to recall. The timed pass is where it becomes usable under pressure.

Implementation checklist

• Restate the input and output before coding.
• Write down the invariant or decision rule in one sentence.
• Test the smallest case, the largest obvious case, and the case that breaks the naive approach.
• Compare the final complexity to the constraint that actually matters.

The checklist is intentionally short. A long checklist becomes another thing to memorize. A short one becomes a rhythm you can run in a blank editor, in a live interview, or in an AlgoArena battle.

Edge cases worth drilling

Most missed solutions come from one of three places: empty input, duplicated values, or a boundary that looks harmless until the loop reaches it. When you review a solution, do not only ask whether it passed. Ask which boundary made the implementation honest.

If a solution uses indexes, trace the first and last iteration. If it uses a map or set, trace the duplicate case. If it uses recursion, name the base case and the state that gets smaller. These small reviews are faster than solving a new problem and often teach more.

Review drill

Before you move on, turn the idea into a small review drill. Pick one representative problem and write three things before you code: the pattern name, the data structure you expect to use, and the edge case that would make the naive version fail. After you solve it, rewrite the explanation in two sentences without looking at the code.

That last step is the difference between recognizing a solution and owning it. If you can explain why the approach works, you can usually rebuild it under pressure. If you only remember the final code shape, the next variant will feel like a new problem.

On AlgoArena, pair this with a timed rep only after the slow explanation is clean. Speed is useful when it compresses a skill you already understand. It is noisy when it hides the fact that the invariant was never clear.

One-session exercise

End with one mixed rep. Choose a problem where the pattern is present but not named in the title. Spend five minutes identifying the signal, ten minutes coding, and five minutes writing a post-solve note about what made the pattern recognizable. If you cannot name the signal before coding, the next best rep is not a harder problem. It is another example of the same pattern with different surface details.

That small debrief is what turns a guide into skill. You are training yourself to notice the shape before the solution is obvious.

Repeat the same exercise once with notes open and once with notes closed. The gap between those two attempts tells you whether the idea is still borrowed or has become part of your own toolkit.

If the closed-notes pass fails, keep the problem but shrink the scope: trace one loop, one recursive call, or one state transition until the reason is obvious.