Instance Storage

Instance storage, also called ephemeral storage, is disk physically attached to the host server running an EC2 instance. It is block-level storage, like EBS, but it is part of the host rather than a network-attached service.

Key characteristics

1. It is temporary. Data survives a reboot but is lost when the instance stops, hibernates or terminates, and is lost if the underlying host fails.
2. It delivers very high IOPS at the lowest latency available, because there is no network between the instance and the disk.
3. It is included in the instance price — there is no separate per-GB charge.
4. Its size and number of devices are fixed by the instance type, and cannot be added after launch.
5. Not every instance type offers it.

Typical uses are caches, buffers, scratch space for a processing step, replicated data that another node also holds, and temporary files that can be regenerated. Anything that must survive a stop belongs on EBS or in a managed service.

Instance store compared with EBS

	Instance store	EBS
Persistence	Lost on stop, hibernate or terminate	Independent of the instance’s life
Latency	Lowest — physically attached	Low — network attached
Maximum IOPS	Very high, fixed by instance type	Up to 80,000 (gp3) or 256,000 (io2 Block Express)
Capacity	Fixed by instance type	1 GiB to 64 TiB per volume, resizable in place
Backup	Nothing built in; copy the data elsewhere	Snapshots, Data Lifecycle Manager, AWS Backup
Redundancy	None — a device failure loses the data	Replicated within the Availability Zone
Cost	Included in the instance price	Charged per GB-month, plus provisioned performance
Use cases	Cache, scratch, buffers, replicated shards	Boot volumes, databases, anything that must persist

The trade is straightforward: instance store buys latency and throughput at the price of every durability guarantee. Designs that use it well treat the local disk as a performance layer over data whose real home is elsewhere.