Home/Learn/System Design/Horizontal vs Vertical Scaling

Horizontal vs Vertical Scaling

Beginner

Fundamentals

Vertical scaling adds more power (CPU/RAM) to a single machine; horizontal scaling adds more machines behind a load balancer. Most large-scale systems rely on horizontal scaling for elasticity and fault tolerance.

Overview

Vertical scaling (scale up) means upgrading a single server with more CPU, RAM, or faster disks. It is simple — no code changes — but has hard limits and a single point of failure. Horizontal scaling (scale out) means adding more machines to a pool and distributing the workload across them using load balancers or partitioning. Horizontal scaling offers near-linear capacity growth, better fault tolerance, and the ability to use commodity hardware. However, it introduces complexity: you need stateless application design (or sticky sessions), distributed data management, and service discovery. In practice, modern systems use horizontal scaling for application tiers and vertical scaling selectively for databases until sharding becomes necessary.

Vertical Scaling

Scale up by adding resources to one machine. Simple but limited — there is a ceiling on how much CPU/RAM a single server can have, and any hardware failure takes the whole service down.

Conceptual — vertical scaling trade-offs

// Vertical scaling example
// Before: 4 CPU cores, 16 GB RAM → handles 1,000 req/s
// After:  32 CPU cores, 128 GB RAM → handles ~6,000 req/s

// Pros:
//  - Zero code changes
//  - No distributed-systems complexity
//  - Single data copy (no consistency issues)

// Cons:
//  - Hardware ceiling (biggest EC2 = 448 vCPU / 24 TB RAM)
//  - Single point of failure
//  - Downtime for upgrades (usually)
//  - Cost grows faster than linearly

Horizontal Scaling

Scale out by adding more servers behind a load balancer. Applications must be stateless — session state moves to a shared store (Redis, DB). This is the foundation of cloud-native design.

Architecture + YAML — horizontal auto-scaling

// Horizontal scaling architecture
//
//  Clients
//    │
//    ▼
// ┌──────────────┐
// │ Load Balancer │  (round-robin, least-connections, IP hash)
// └──┬───┬───┬───┘
//    │   │   │
//    ▼   ▼   ▼
//  App  App  App    ← stateless instances (share nothing)
//    │   │   │
//    ▼   ▼   ▼
// ┌──────────────┐
// │ Shared State  │  Redis / DB / S3
// └──────────────┘

// Kubernetes Horizontal Pod Autoscaler
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: order-service-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: order-service
  minReplicas: 3
  maxReplicas: 20
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70

When to Choose Which

Start vertical for simplicity (especially databases). Go horizontal when you hit hardware limits, need fault tolerance, or want elastic scaling. Most production systems are horizontally scaled at the app tier with vertically scaled (or sharded) databases.

Conceptual — decision matrix

// Decision matrix
//
// Factor              | Vertical       | Horizontal
// ───────────────────────────────────────────────────
// Complexity          | Low            | High
// Cost efficiency     | Moderate       | Better at scale
// Fault tolerance     | Single point   | Redundant
// Max capacity        | Hardware limit | ~Unlimited
// Stateful workloads  | Easier         | Needs shared state
// Downtime for scale  | Usually yes    | Zero-downtime
//
// Real-world combo:
// Netflix: horizontally scaled microservices + vertically scaled Cassandra nodes
// Uber:    horizontally scaled services + sharded MySQL (Schemaless)

Key Points to Remember

1Vertical scaling = bigger machine; horizontal scaling = more machines.
2Horizontal scaling requires stateless app design — push state to Redis, DB, or object storage.
3Most cloud-native systems scale horizontally at the app tier and use database sharding when vertical limits are hit.
4Kubernetes HPA automates horizontal scaling based on CPU, memory, or custom metrics.
5Start simple (vertical) and evolve to horizontal as traffic demands grow.

Interview Questions

What is the difference between horizontal and vertical scaling?

EasyTCS

Why do microservices favour horizontal scaling?

EasyInfosys

How would you scale a stateful service horizontally?

MediumAmazon

What are the limits of vertical scaling in cloud environments?

MediumGoogle

Design an auto-scaling strategy for a flash-sale e-commerce backend.

HardFlipkart

Ask Aria about Horizontal vs Vertical Scaling

Your personal AI tutor — ask anything about this concept

Revision Status

Personal Notes

Discussion

Loading discussion…

All System Design Concepts

CAP Theorem