Block storage

Storage exposed as raw blocks attached to a VM like a disk, supporting random read/write and filesystems. EBS is the canonical cloud example. Low latency, persistent, but expensive and not internet-addressable.

When to use

Database storage, VM boot disks, anything needing a filesystem and random I/O.
Low-latency persistent state.

When not to

Large immutable blobs (use object storage).
Shared concurrent writes from many nodes.

Definition

Block storage presents fixed-size blocks over a network (iSCSI, NVMe-oF) to an OS, which formats it with a filesystem. It supports random reads/writes and is the basis for databases and boot disks.

Why it exists

Databases and OSes need random-access, low-latency, filesystem-backed storage. Object storage can't do that. Block storage gives a VM a disk that performs like local storage but is network-attached and snapshottable.

How it works

A volume is attached to one instance at a time (usually). Reads/writes go over the storage network. IOPS and throughput scale with volume size/type (provisioned IOPS SSDs vs gp). Snapshots are block-incremental.

Scaling limits

Attached to one instance (no shared multi-writer easily). Performance is bounded by volume type and network. Cost scales with provisioned capacity/IOPS, not just used space.

Cost shape

Per-GB + provisioned IOPS/throughput. Blow-ups: over-provisioned IOPS, large idle volumes, and snapshot storage.

Alternatives

Object storage — Cheap and scalable, but no random writes or filesystem.
Ephemeral instance storage — Fastest but lost on instance stop.

Failure modes

IOPS limits throttling DBs.
Single-attach limiting HA.
Snapshot restore time.

Object storage, WAL (write-ahead log), Snapshot, Replication