EC2 Instance Store
EC2 Instance Store is the ultimate "speed demon" of the AWS storage world. If EBS is like plugging in a network-attached USB stick, an Instance Store is like soldering a blazing-fast NVMe SSD directly onto the physical motherboard of the server host.
Key Takeaways
The Core Performance Advantage
- Physical Attachment: Unlike EBS (which sends data back and forth over a network wire), an instance store is physically connected to the underlying hardware server hosting your virtual machine.
- Raw Power It delivers insanely high I/O performance. Stephane points out that an Instance Store can hit millions of IOPS (e.g., 3.3 million reads), while a standard
gp2EBS volume maxes out around 32,000 IOPS.
| Instance type | Instance store volumes | Instance store type | 100% random read IOPS / Write IOPS |
|---|---|---|---|
| i3.large | 1 x 475 GB | NVMe SSD | 103,125 / 35,000 |
| i3.xlarge | 1 x 950 GB | NVMe SSD | 206,250 / 70,000 |
| i3.2xlarge | 1 x 1900 GB | NVMe SSD | 412,500 / 180,000 |
| i3.4xlarge | 2 x 1900 GB | NVMe SSD | 825,000 / 360,000 |
| i3.8xlarge | 4 x 1900 GB | NVMe SSD | 1,650,000 / 720,000 |
| i3.16xlarge | 8 x 1900 GB | NVMe SSD | 3,300,000 / 1,440,000 |
| i3.metal | 8 x 1900 GB | NVMe SSD | 3,300,000 / 1,440,000 |
| i3en.large | 1 x 1250 GB | NVMe SSD | 42,500 / 32,500 |
| i3en.xlarge | 1 x 2500 GB | NVMe SSD | 85,000 / 65,000 |
| i3en.2xlarge | 2 x 2500 GB | NVMe SSD | 170,000 / 130,000 |
| i3en.3xlarge | 1 x 7500 GB | NVMe SSD | 250,000 / 200,000 |
| i3en.6xlarge | 2 x 7500 GB | NVMe SSD | 500,000 / 400,000 |
| i3en.12xlarge | 4 x 7500 GB | NVMe SSD | 1,000,000 / 800,000 |
| i3en.24xlarge | 8 x 7500 GB | NVMe SSD | 2,000,000 / 1,600,000 |
| i3en.metal | 8 x 7500 GB | NVMe SSD | 2,000,000 / 1,600,000 |
Subset of i3 and i3en instance types with instance store volumes and their performance specs. For the full list, see AWS Documentation.
The Catch: Ephemeral Storage
- The Wipe Rule: This storage is strictly ephemeral (temporary).
- The Danger Zone: If you STOP or TERMINATE the instance, or if the underlying physical hardware suffers a hardware failure, the data is gone forever. You cannot get it back. -Survival Mode: The data only survives a standard OS reboot.
Ideal Use Cases
Because it's fast but fragile, you only use it for data that can be lost without causing a production disaster.
- High-performance caches and buffers.
- Scratch space or temporary file processing directories.
- Distributed file systems or databases that handle replication at the software level (where losing one node doesn't kill the cluster).
Exam Tips
The "Stop" Limitation: For EC2 instances backed by only by an Instance Store, **the "Stop" action is completely disabled in the AWS Console/CLI". You can only "Reboot" or "Terminate" them. Why? Because stopping them means moving them to a different physical host, which would wipe the drive anyway.
Keyword Association: On the exam, if you see phrases like maximum I/O performance, Lowest Latency, local storage, or "non-persistent data", your mind should immediately skip EBS and choose Instance Store.
The Backup Responsibility: If a question asks who is responsible for backing up instance store data, it 100% on the developer. AWS will not snapshot or replicate this data for you. You have to write scripts to copy it to S3 or EBS manually.