EC2 & AWS Compute Services¶

From AWS Apprenticeship notes — the full compute service landscape.

Ultra-Short Summary¶

EC2 is AWS's virtual machine service — you pick the OS, instance type, and configuration. But EC2 is just one compute option. AWS has a full spectrum from bare metal to fully serverless, and choosing the right one depends on how much control vs simplicity you need.

The Compute Spectrum¶

Most control, most management                    Least control, most managed
        ↓                                                    ↓
Bare Metal → EC2 → ECS (containers) → Lambda (serverless)
                         ↓
                      Fargate
                 (serverless containers)

All AWS Compute Services¶

Service	What It Is	When to Use
EC2	Virtual machines — full OS control	Long-running workloads, custom OS config, legacy lift-and-shift
EC2 Auto Scaling Groups	Automatically adjust EC2 count	Variable traffic — scale up/down based on demand
ECS	Container orchestration (Docker on AWS)	Containerised microservices without needing Kubernetes
EKS	Managed Kubernetes control plane	Need K8s specifically — team expertise or portability
Fargate	Serverless compute for containers	Container benefits without managing EC2/clusters
Lambda	Serverless functions, pay per request	Event-driven, short-lived tasks, automation
Batch	Managed batch job processing	Compute-heavy, scheduled, queue-based workloads
Lightsail	Simplified low-cost VPS	Simple apps, personal projects, WordPress
Elastic Beanstalk	PaaS — upload code, AWS handles infra	Fast app deployment without DevOps expertise
App Runner	Fully managed web app/API hosting	Push repo → auto-deploys, zero config
ECR	Private Docker image registry	Store and version container images for ECS/EKS/Fargate
Outposts	AWS hardware physically on-prem	Ultra-low latency or strict data residency requirements

EC2 Instance Types — The Pattern¶

Instance type names follow a pattern: <family><generation>.<size>

m5.xlarge
│ │  └── size: nano, micro, small, medium, large, xlarge, 2xlarge...
│ └──── generation (5 = 5th gen)
└────── family (m = general purpose)

Instance Families¶

Family	Optimised For	Example Use Case
`m`	General purpose (balanced)	Web servers, app servers
`c`	Compute (high CPU)	Gaming, video encoding, ML inference
`r`	Memory (high RAM)	In-memory DB, real-time processing
`t`	Burstable (cheap, shared CPU)	Dev/test, low-traffic apps
`i`	Storage (NVMe instance store)	High I/O databases
`g` / `p`	GPU	ML training, graphics rendering
`d`	Dense storage	Data warehousing
`hpc`	High Performance Computing	Scientific simulation

Graviton instances add g suffix to family: m7g, c7g — AWS-built ARM processors, 20-40% better price/performance.

EC2 Pricing Models¶

Model	How It Works	Best For	Discount vs On-Demand
On-Demand	Pay by hour/second, no commitment	Unpredictable workloads, dev/test	—
Reserved	1 or 3 year commitment	Steady-state, predictable workloads	Up to 72%
Savings Plans	Flexible commitment ($/hour for 1-3yr)	Mix of instance types/regions	Up to 66%
Spot	Bid on unused capacity, can be interrupted	Fault-tolerant batch, ML training	Up to 90%
Dedicated Host	Physical server dedicated to you	Compliance, bring-your-own-license	—

SAA Pattern

"Steady state + save money" → Reserved or Savings Plans. "Can tolerate interruption" → Spot. "Unpredictable traffic" → On-Demand + Auto Scaling.

EC2 Key Decisions at Launch¶

AMI (Amazon Machine Image)¶

The blueprint for your instance — OS, pre-installed software, initial state.

AWS-provided: Amazon Linux 2023, Ubuntu, Windows Server
AWS Marketplace: pre-built software stacks
Custom AMI: your own image (useful for fast Auto Scaling — no bootstrap delay)

Storage¶

Type	Persists?	Speed	Use Case
EBS (root volume)	Yes	Fast	OS, application data
EBS (additional)	Yes	Fast	Database, logs
Instance Store	No (ephemeral)	Very fast	Scratch, cache, temp

Networking¶

VPC + Subnet — which network to place the instance in
Security Group — stateful firewall (what traffic is allowed in/out)
Public IP — optional, assigned at launch (changes on stop/start)
Elastic IP — static public IP, persists across stop/start
Key Pair — SSH access (.pem file)

User Data — Bootstrap Scripts¶

Run commands automatically when an EC2 instance first starts:

#!/bin/bash
yum update -y
yum install -y httpd
systemctl start httpd
systemctl enable httpd
echo "<h1>Hello from $(hostname)</h1>" > /var/www/html/index.html

Paste this into "User Data" at launch. Useful for: - Installing software - Pulling application code - Configuring services

EC2 Instance Metadata¶

From inside any EC2 instance, you can query metadata about itself:

# IMDSv2 (current standard)
TOKEN=$(curl -X PUT "http://169.254.169.254/latest/api/token" \
  -H "X-aws-ec2-metadata-token-ttl-seconds: 21600")

curl -H "X-aws-ec2-metadata-token: $TOKEN" \
  http://169.254.169.254/latest/meta-data/instance-id

curl -H "X-aws-ec2-metadata-token: $TOKEN" \
  http://169.254.169.254/latest/meta-data/public-ipv4

curl -H "X-aws-ec2-metadata-token: $TOKEN" \
  http://169.254.169.254/latest/meta-data/iam/security-credentials/

169.254.169.254 is a link-local address — only reachable from within the EC2 instance.

Security Note

The last endpoint above returns temporary IAM credentials for the attached role. This is the endpoint exploited in SSRF attacks. Always enforce IMDSv2 and least-privilege IAM.

Mental Model¶

Need a server?
        ↓
Do you need the full OS / custom software?
  Yes → EC2
  No  ↓
Do you need containers?
  Yes → ECS Fargate (no servers) or ECS EC2 (you manage servers)
  No  ↓
Is it event-driven and short-lived?
  Yes → Lambda
  No  ↓
Is it a web app / API you want to deploy fast?
  Yes → App Runner or Elastic Beanstalk

SAA Patterns¶

Scenario	Answer
Unpredictable traffic, scale automatically	EC2 + Auto Scaling Group + ALB
Long-running batch job, can tolerate interruption	EC2 Spot
Steady web server, minimise cost	Reserved Instance or Savings Plan
Containerised microservice, no server management	ECS Fargate
Simple CRUD API, event-triggered	Lambda
Legacy app needs full OS and specific libraries	EC2
Fast deployment of web app from GitHub	App Runner

30-Second Takeaway¶

EC2 = full VM, you manage the OS. Fargate = serverless containers. Lambda = serverless functions.
Instance type pattern: family + generation + size (e.g. m5.xlarge)
t instances are cheap and burstable. c is compute. r is memory. m is balanced.
Spot = cheapest, can be interrupted. Reserved = commit for discount. On-Demand = flexible.
Instance metadata at 169.254.169.254 — enforce IMDSv2.

Self-Quiz¶

What does the instance type c5.2xlarge tell you about the instance?
When would you choose Fargate over EC2 for running containers?
What's the difference between Reserved Instances and Savings Plans?
Your app has traffic spikes. What EC2 feature handles automatic scaling?
What is an AMI and when would you create a custom one?
What's at 169.254.169.254 and why is it a security concern?
An EC2 instance needs to access S3 without hardcoded keys. How?
What instance family would you choose for an in-memory database?