☁︎SAA-C03

Instance Store

Instance Store — Concept

What it is

Instance Store = physical disk(s) (NVMe SSD or HDD) directly attached to the host of certain EC2 families. Provides the highest possible IOPS / lowest latency but is ephemeral.

Why it exists

Network-attached EBS adds latency; some workloads (high-perf cache, big-data scratch, NoSQL temp store) need millions of IOPS at sub-100 µs.

How it works

  • Storage is tied to the host hardware, not the instance.
  • Data is lost when:
    • The instance is stopped or terminated
    • The host fails (hardware retired)
    • The instance is hibernated to a non-EBS root (rare)
  • Data survives a reboot of the instance.
  • Cannot be detached or moved like EBS.
  • Cannot be encrypted by you (newer Nitro instances are encrypted with host-managed keys).

When you get it

  • Only certain families: i3, i3en, i4i, d2, d3, h1, some c5d, m5d, r5d, c6gd, g4dn, x1, …
  • Size is fixed by the instance type (e.g. i3.16xlarge = 8 × 1.9 TB NVMe).
  • Free — included in instance hourly cost.

When to use vs alternatives

Use ...Instead of ...When ...
Instance StoreEBSYou need millions of IOPS / sub-100 µs latency, can rebuild from replica
EBSInstance StoreData must persist across stop/terminate
Instance Store + EBS rootAll EBSMix: ephemeral scratch on instance store, OS + persistent on EBS

Common exam scenarios

  1. "Cassandra / NoSQL cluster, replicas, highest IOPS, lowest cost" → i3 / i4i Instance Store (data sharded across nodes).
  2. "Big data scratch / temp processing" → Instance Store on d3 / i3en.
  3. "App stores user uploads, needs to persist after stop" → NOT Instance Store; use EBS or S3.
  4. "What happens to data when EC2 stops?" → Instance Store data is lost, EBS data persists.

Exam tip

The keyword "ephemeral" or "can be lost" = Instance Store. If the question mentions persistence, snapshots, or moving between instances → it's EBS.

References