AWS SA - Associate EC2 & Storage

EC2

Virtual machines on demand in AWS. Pay per hour/second while running.

AMI

Template to launch EC2 instances. Prebaking = software pre-installed for faster launch.

User Data

Script that runs automatically at first boot. Exam clue: automate setup at launch.

EC2 Hibernate

Saves RAM to EBS root volume. Resumes exactly where it left off.

Instance Families

General (M/T), Compute (C), Memory (R/X), Storage (I), GPU (P/G), HPC.

Key Pairs

SSH (Linux) or RDP (Windows) authentication. Regional. Private key = your responsibility.

Session Manager

Preferred EC2 access. No open ports needed. IAM controlled. Logs everything.

Bastion Host

Jump server in public subnet to reach private instances. Less secure than SSM.

IMDS

Instance metadata at 169.254.169.254. Always same IP. Use IMDSv2 (token-based).

On-Demand

Pay per hour. No commitment. Most expensive. Use for unpredictable workloads.

Reserved Instances

1 or 3 year commit. Up to 72% off. Best for steady predictable workloads.

Savings Plans

Flexible RI alternative. Commit to spend per hour. Applies to EC2/Lambda/Fargate.

Spot Instances

Up to 90% off. Can be interrupted anytime. Never for critical workloads.

Dedicated Host

Physical server for you only. Full control. BYOL/compliance. Most expensive.

Capacity Reservation

Guaranteed capacity in specific AZ. No time commitment. Charged even if unused.

EBS

Virtual hard drive. Network attached. AZ-level. One instance at a time (usually).

EBS vs Instance Store

EBS = persistent network storage. Instance Store = fast but lost on stop/terminate.

EBS Volume Types

gp3 = default general. io2 = high IOPS databases. st1 = throughput. sc1 = cold cheapest.

EBS Snapshots

Incremental backups stored in S3. Copy across regions/accounts. First snapshot = full.

EBS Encryption

Must enable at creation. To encrypt existing = snapshot + encrypted copy + restore.

EBS Multi-Attach

io1/io2 only. Up to 16 instances. Same AZ. Needs cluster-aware file system.

Delete on Termination

Root volume deleted by default. Additional volumes kept by default.

Data Lifecycle Manager

Automates EBS snapshot creation and retention on a schedule. No scripting needed.

Launch Templates

Versioned EC2 launch configs. Supports multiple versions. Replaces launch configurations.

EFS

Shared Linux file system. Multiple EC2 across multiple AZs. Auto-scales. NFSv4. Pay per use.

EFS vs EBS

EFS = shared across many instances/AZs. EBS = dedicated to one instance in one AZ.

FSx for Windows

SMB/NTFS Windows file system. Active Directory integration. For Windows apps.

FSx for Lustre

HPC/ML/big data. Massive throughput. Integrates with S3 as backend.

FSx for ONTAP

Managed NetApp in AWS. For migrating from on-premises NetApp environments.

FSx for OpenZFS

Managed ZFS. NFS protocol. Multi-OS support.

Instance Store

Physically attached. Extremely fast. Lost on stop/terminate. Temp data only.

Placement Groups

Cluster = low latency same AZ. Spread = different hardware HA. Partition = large distributed.

Instance families to know

M (general purpose), C (compute), R (memory), I (storage/NVMe), P/G (GPU)

gp3 vs gp2

gp3 has fixed baseline IOPS (3000) independent of size, cheaper — default choice now

io1/io2 vs gp3

Use io1/io2 only when workload needs high, consistent IOPS (databases), not general purpose

st1 vs sc1

st1 = big sequential throughput (log processing, big data); sc1 = cheapest, infrequent access

Cluster placement group

Same AZ, packed close together — for low latency/high throughput (HPC, tightly coupled apps)

Spread placement group

Different hardware — for resilience, minimizes correlated failure (max 7 instances per AZ)

Partition placement group

Logical partitions on separate hardware — for large distributed systems (Hadoop, Cassandra, Kafka)