AWS for DevOps — ECS: Fargate vs EC2

7.1 — ECS: Fargate vs EC2 Launch Types

Amazon Elastic Container Service (ECS) is AWS’s container orchestrator. When you run containers on ECS, you choose a launch type — this determines who manages the underlying compute.

The Two Launch Types

  ECS Cluster
  ┌──────────────────────────────────────┐
  │                                       │
  │  Fargate Launch Type                  │
  │  ┌────────┐ ┌────────┐ ┌────────┐   │
  │  │ Task A │ │ Task B │ │ Task C │   │
  │  │(0.5 CPU│ │(1 CPU  │ │(2 CPU  │   │
  │  │ 1GB)   │ │ 2GB)   │ │ 4GB)   │   │
  │  └────────┘ └────────┘ └────────┘   │
  │  AWS manages the servers ▲           │
  │                                       │
  │  EC2 Launch Type                      │
  │  ┌─── EC2 Instance ───┐              │
  │  │ ┌──────┐ ┌──────┐  │ ┌─── EC2 ──┐│
  │  │ │Task D│ │Task E│  │ │ ┌──────┐ ││
  │  │ └──────┘ └──────┘  │ │ │Task F│ ││
  │  │   You manage this ▲ │ │ └──────┘ ││
  │  └────────────────────┘ └──────────┘│
  └──────────────────────────────────────┘

Comparison

Aspect	Fargate	EC2
Server management	None — AWS handles it	You manage EC2 instances
Scaling	Per-task (instant)	Scale instances + tasks
Pricing	Per vCPU/GB per second	EC2 instance cost (can use Reserved/Spot)
Cost at scale	More expensive	Cheaper with Reserved Instances
GPU support	No	Yes
Docker-in-Docker	No	Yes (privileged mode)
EBS volumes	No (EFS only)	Yes
SSH into container host	No	Yes
Startup time	~30-60 seconds	Faster (instances already running)
Best for	Most workloads, teams wanting simplicity	GPU workloads, cost optimisation at scale

Creating an ECS Cluster

# Fargate-only cluster (no EC2 instances to manage)
$ aws ecs create-cluster --cluster-name my-app-cluster
{
    "cluster": {
        "clusterArn": "arn:aws:ecs:ap-south-1:123456789012:cluster/my-app-cluster",
        "clusterName": "my-app-cluster",
        "status": "ACTIVE"
    }
}

# Cluster with EC2 capacity provider
$ aws ecs create-cluster \
    --cluster-name my-ec2-cluster \
    --capacity-providers FARGATE FARGATE_SPOT \
    --default-capacity-provider-strategy \
        capacityProvider=FARGATE,weight=1 \
        capacityProvider=FARGATE_SPOT,weight=3

Capacity Providers

Capacity providers control where tasks run and how they scale:

Provider	Description	Use Case
FARGATE	Standard Fargate	Production workloads
FARGATE_SPOT	Spot pricing (up to 70% cheaper)	Fault-tolerant tasks (batch, dev)
EC2 Auto Scaling Group	Your own instances	GPU, cost optimisation

# Mix Fargate and Fargate Spot (75% Spot, 25% On-Demand)
$ aws ecs put-cluster-capacity-providers \
    --cluster my-app-cluster \
    --capacity-providers FARGATE FARGATE_SPOT \
    --default-capacity-provider-strategy \
        capacityProvider=FARGATE,weight=1,base=1 \
        capacityProvider=FARGATE_SPOT,weight=3

The base=1 ensures at least 1 task always runs on standard Fargate (reliable), while additional tasks prefer Spot (cheaper).

When to Choose What

  "I just want to run containers"
       │
    Fargate
       │
  "I need GPUs / Docker-in-Docker / SSH access"
       │
    EC2
       │
  "I want cheaper Fargate for non-critical tasks"
       │
    Fargate Spot
       │
  "I have 100+ containers and want to save money"
       │
    EC2 with Reserved Instances
       │
  "I already use Kubernetes"
       │
    EKS (not ECS)

Real-World Example

A typical production setup:

# Production: Standard Fargate (reliable)
$ aws ecs create-service \
    --cluster production \
    --service-name api \
    --launch-type FARGATE \
    --task-definition api:5 \
    --desired-count 3

# Development: Fargate Spot (cost-effective)
$ aws ecs create-service \
    --cluster development \
    --service-name api \
    --capacity-provider-strategy \
        capacityProvider=FARGATE_SPOT,weight=1 \
    --task-definition api:5 \
    --desired-count 1

Common Mistakes

Mistake	Why It Hurts	Fix
Using Fargate for GPU workloads	Fargate doesn’t support GPUs	Use EC2 launch type with GPU instances
Fargate Spot for critical prod services	Tasks can be interrupted	Use standard Fargate for production
Over-provisioning Fargate tasks	Paying for unused CPU/memory	Right-size using Container Insights metrics
Not using capacity providers	Stuck on one launch type	Configure mixed strategies

Quick Reference

Task	Command
Create cluster	aws ecs create-cluster --cluster-name NAME
List clusters	aws ecs list-clusters
Describe cluster	aws ecs describe-clusters --clusters NAME
Delete cluster	aws ecs delete-cluster --cluster NAME
Set capacity providers	aws ecs put-cluster-capacity-providers --cluster NAME ...

---

Next → ECR — Elastic Container Registry