Load Balancing

Introduction

Load balancers distribute traffic across servers for throughput and fault tolerance. Layer 4 (TCP) vs Layer 7 (HTTP) changes what you can route on.

The story

A launch day spike hit one app server because DNS pointed directly at it. Adding an ALB with health checks and round-robin spread load; unhealthy nodes drained automatically. Uptime went from "hope" to infrastructure.

The business problem

Teams that skip disciplined Load Balancing thinking pay in outages, cost overruns, and failed interviews:

• Outages: components chosen without scale math fail at peak.
• Cost: over-engineered microservices for 100-user products.
• Latency: missing cache/CDN/replica on read-heavy paths.
• Interviews: boxes without numbers and trade-offs don't hire.

The problem teams faced

This lesson addresses:

• When and why Load Balancing matters in real architectures.
• How to sketch components and data flows under interview time pressure.
• Trade-offs vs alternatives — what senior engineers articulate aloud.
• Production patterns and failure modes you've seen or will see.

Understanding the topic

Core idea: Load Balancing in production system design.

• Problem — what force this topic addresses.
• Building blocks — components typically involved.
• Trade-offs — what you gain and what you pay.
• Interview — how this appears in design loops.

Internal architecture

Load Balancing — system view:

Clients / CDN
   ↓
Load Balancer (L4 or L7)
   ↓ health checks
[App Server 1 | App Server 2 | App Server N]
   ↓
Shared session store (if stateful) or stateless JWT

Visual explanation

Three diagrams: architecture flow, design process, and scaling lens:

Informative example

Example — Load Balancing:

L4 LB: fast, IP/port routing — good for TCP databases, gaming
L7 LB: HTTP path routing — /api/* → API fleet, /static/* → CDN
Algorithms: round-robin, least connections, consistent hash (sticky sessions)

Execution workflow

Real-world use

Used in production at major tech companies and every FAANG system design interview loop. Patterns align with AWS/GCP well-architected frameworks and Google SRE practice.

Production case study

A launch day spike hit one app server because DNS pointed directly at it. Adding an ALB with health checks and round-robin spread load; unhealthy nodes drained automatically. Uptime went from "hope" t…

• Context: production or interview scenario from this lesson.
• Decision: component and trade-off choices explained.
• Outcome: measurable latency, availability, or cost impact.

Trade-offs

• Pro: structured approach reduces outages and interview failures.
• Con: upfront thinking takes time — faster than wrong rebuild.
• Con: every product has unique constraints — adapt templates.

Decision framework

• Always estimate before drawing microservices.
• Match consistency model to business domain (money vs likes).
• Document rejected alternatives — ADR or interview closing.

Best practices

• State assumptions explicitly (DAU, read:write ratio).
• Draw async vs sync paths with different line styles.
• Close with monitoring and on-call failure modes.

Anti-patterns to avoid

• Jumping to Kafka and microservices without scale justification.
• Single DB box with no read replica or cache on read-heavy design.
• Ignoring idempotency on write APIs with retries.

Common mistakes

• Underestimating peak QPS (forgetting peak factor).
• Hot shard from poor partition key choice.

Debugging tips

• Trace one request ID through diagram — find missing component.
• Ask "what fails first at 10×?" for every design.

Optimization strategies

• Cache + CDN first for read-heavy; shard when single DB saturates.
• Async queue for slow side effects (email, analytics).

Common misconceptions

• More components ≠ better design — simplest meeting NFRs wins.
• Interviews test process — not memorizing Netflix architecture.

Advanced interview questions

Summary

You can explain Load Balancing in a 45-minute system design interview with numbers, diagrams, and trade-offs. Teach it back without notes — you own this piece.