Mohak Gupta@rushbeef04

Load Balancing, Explained Like Managing a Restaurant

When an application grows, a single server cannot handle all users. So we create multiple instances of the same service.

Now the key question: Which instance should handle a new request?

This is where a Load Balancer (LB) comes in.

Think of it like a restaurant manager assigning customers to different tables so that no waiter is overloaded and service stays smooth.

load balancer

Types of Load Balancers

The difference lies in how much of the request they understand.

1. L4 Load Balancer (Transport Layer)

  • Works at connection level
  • Uses IP address and port
  • Operates over TCP/UDP
  • Does not inspect request content

Analogy: A security guard who only checks your ID and lets you pass, without asking your purpose.

2. L7 Load Balancer (Application Layer)

  • Works at request level
  • Understands HTTP/HTTPS/WebSockets
  • Can inspect headers, cookies, paths, body
  • Routes based on content

Analogy: A receptionist who asks why you’re here and sends you to the correct department.

L4 Load Balancer Modes

1. Passthrough Mode (Most Common)

  • Does not break the TCP connection
  • Simply forwards packets to backend

Workflow:

Client → LB → Server
Server → LB → Client

The client thinks it is directly talking to the server.

Used when: performance and speed are more important than control.

2. Proxy Mode

  • Breaks the client connection
  • Creates a new connection to backend
  • More control over traffic

Analogy: A middleman who talks to both sides separately.

Used when: you need control, logging, filtering, or security rules.

Load Balancing Algorithms

Now comes the core logic: How does the load balancer decide where to send a request?

These algorithms fall into two categories:

  • Static → fixed rules
  • Dynamic → real-time decisions

1. Round Robin

Requests are distributed sequentially across servers.

Example:

Request 1 → A
Request 2 → B
Request 3 → C
Request 4 → A

Use case: Simple systems where all servers are equal.

Limitation: Does not consider server capacity or current load.

Incoming Requests
      ↓
A ← B ← C ← A ← B ← C
(Equal distribution)

round robin

2. Weighted Round Robin

Each server is assigned a weight based on capacity.

Example:

  • A (weight 3), B (weight 1)
A → A → A → B → repeat

Use case: When servers have different power.

Limitation: Still ignores actual workload. A heavy request may hit a weaker server.

A (High capacity) → more requests
B (Low capacity)  → fewer requests
But request complexity is ignored

weighted round robin

3. IP Hash

Maps a user’s IP to a specific server using hashing.

Use case: Session persistence (same user always hits same server)

Analogy: A bodyguard checks your ID and always sends you to the same room.

User IP → Hash Function → Server
Same IP → Same Server

ip hash

4. Least Connections

Routes request to server with fewest active connections.

Example:

  • A → 10 connections
  • B → 3 connections

New request → B

Use case: Dynamic traffic environments.

Limitation: Does not consider server strength.

Server A → 10 active
Server B → 3 active  ← chosen

least conn

5. Weighted Least Connections

Considers both load and capacity.

Formula:

Connections / Weight

Example:

  • Strong server → 2 connections, weight 10 → 0.2
  • Weak server → 1 connection, weight 1 → 1

Request goes to strong server

Use case: Real-world systems with uneven infrastructure.

Server A → 2/10 = 0.2  ← chosen
Server B → 1/1  = 1

weighted least conn

6. Least Response Time

Chooses the fastest server based on response speed.

Metric used: TTFB (Time To First Byte)

Formula:

TTFB × Active Connections

Example:

  • A → 3 × 2 = 6
  • B → 2 × 0 = 0

Request goes to B

If equal → fallback to Round Robin

Use case: Performance-critical systems.

(diagram)

Server A → slower (6)
Server B → faster (0) ← chosen

Conclusion

Load balancing is the practice of distributing traffic intelligently to keep systems fast, stable, and available.

L4 load balancers focus on speed and simplicity, while L7 load balancers provide deeper control and smarter routing. Similarly, static algorithms are easy to implement, but dynamic algorithms adapt better to real-world traffic.

There is no one-size-fits-all solution. The right choice depends on system scale, traffic patterns, and performance needs. In practice, modern systems combine multiple strategies to achieve both efficiency and reliability.

At its core, a good load balancer works like a skilled manager — it understands the situation, distributes work wisely, and ensures no single component becomes a bottleneck.

Discussion

Loading comments…