EBS

An EC2 instance’s local disks vanish the moment the instance stops or fails. Amazon EBS (Elastic Block Store) solves this by providing network-attached block volumes that live independently of any instance: you attach a volume to an instance, it appears as a raw block device you can format and mount, and the data survives stop/start, instance replacement, and hardware failure. It’s the storage you reach for whenever you need a real filesystem or a database’s data directory on EC2.

• An EBS volume is AZ-locked — it can only be attached to an instance in the same Availability Zone. It is replicated across multiple devices within that AZ for durability, but the AZ is the blast radius.
• To move a volume to another AZ or Region, you take a snapshot (stored in S3, regional) and create a new volume from it in the target AZ; snapshots can also be copied across Regions.
• Multi-Attach (io1/io2 only) lets one volume attach to multiple instances, but still only within a single AZ.

graph LR
subgraph AZA["Availability Zone A"]
  EC2["EC2 Instance"]
  VOL["EBS Volume (gp3, 100 GiB)"]
  EC2 -->|"attach as /dev/xvdf"| VOL
end
VOL -->|"point-in-time snapshot"| SNAP["EBS Snapshot"]
subgraph S3["Amazon S3 (Regional, AWS-managed)"]
  SNAP
end
SNAP -->|"create volume"| VOLB["EBS Volume in AZ B"]
SNAP -.->|"copy snapshot"| REGION2["Another Region"]
subgraph AZB["Availability Zone B"]
  VOLB
end
classDef vol fill:#e8f0fe,stroke:#3b82f6,stroke-width:2px,color:#173a70
classDef snap fill:#f3ecfb,stroke:#8b3de0,stroke-width:2px,color:#3a1a5c
class VOL,VOLB vol
class SNAP snap

Type	Family	Best for	Key limit
gp3	SSD	General purpose default — boot volumes, most workloads	3,000 IOPS / 125 MB/s baseline, provisioned independently of size
gp2	SSD	Legacy general purpose	IOPS scale with size (3 IOPS/GiB)
io2 Block Express	SSD	Latency-sensitive databases needing sustained high IOPS	Up to 256,000 IOPS, 99.999% durability
st1	HDD	Big, sequential throughput — log processing, data warehouses	Cannot be a boot volume
sc1	HDD	Cold, infrequently accessed data	Lowest cost, cannot be a boot volume

You pay for provisioned capacity per GB-month — whether the volume is 5% full or 95% full — for the entire time the volume exists, even while its instance is stopped. gp3 additionally lets you pay separately for IOPS and throughput above the free baseline. Snapshots bill per GB-month of changed blocks (they’re incremental), stored in S3. Deleting the instance does not delete attached volumes unless “Delete on Termination” is set — orphaned volumes are a classic silent cost leak.

• “Volume won’t attach” is almost always an AZ mismatch — the volume and instance must be in the same AZ. The fix is snapshot → create volume in the right AZ.
• Snapshots are incremental but deletion is safe — AWS keeps whatever blocks later snapshots still need. Restoring from a snapshot is lazy-loaded from S3, so first-touch reads are slow unless you enable Fast Snapshot Restore (extra cost).
• gp3 is the modern default: on gp2 you had to over-provision size just to get IOPS; on gp3 you provision IOPS/throughput directly, usually at ~20% lower cost.
• EBS volumes can be encrypted with KMS; encryption is set at creation. You can’t un-encrypt, but you can snapshot and restore into an encrypted volume. Snapshots of an encrypted volume are encrypted automatically.
• You can resize a volume, change its type, or adjust IOPS live (Elastic Volumes) without detaching — but you must then extend the filesystem inside the OS.
• For RAID: RAID 0 across volumes for more raw IOPS/throughput; never rely on RAID for durability — use snapshots.