AWS Storage
Snow Family
Physical devices for offline petabyte-scale data transfer and edge computing
The AWS Snow Family consists of physical devices shipped to your location for offline data transfer and edge computing, enabling petabyte-scale migrations when network bandwidth would be too slow or expensive. The family ranges from the portable Snowcone (8TB usable) to the datacenter-scale Snowmobile (100 PB), each with varying compute and storage capabilities. Snow devices are also used for edge computing in disconnected or bandwidth-constrained environments like oil rigs, factories, and military deployments.
Snow Family Devices - Comparing Snowcone, Snowball, and Snowmobile
Each Snow device is optimized for a different scale of data transfer and compute need. Choosing the wrong device wastes money on shipping or requires multiple orders.
| Device | Usable Storage | Compute | Network | Use Case |
|---|---|---|---|---|
| Snowcone | 8 TB HDD or 14 TB SSD | 2 vCPU, 4 GB RAM | 10 GbE, WiFi | Edge locations, IoT, small migrations (<8 TB) |
| Snowcone SSD | 14 TB NVMe SSD | 2 vCPU, 4 GB RAM | 10 GbE, WiFi | Same as Snowcone, faster local storage |
| Snowball Edge Storage Optimized | 80 TB usable HDD | 40 vCPU, 80 GB RAM, 1 TB NVMe cache | 25 GbE | Large data migrations, local compute |
| Snowball Edge Compute Optimized | 28 TB usable NVMe SSD | 104 vCPU, 416 GB RAM, optional GPU | 25 GbE, 100 GbE optional | Machine learning inference, video processing at edge |
| Snowmobile | Up to 100 PB | None (storage only) | Multiple 40 GbE connections | Exabyte-scale datacenter migrations |
Rule of thumb: if transferring your data over the internet would take more than a week at your available bandwidth, Snow devices are likely more economical. Use the AWS Snow Family Data Transfer Calculator to compare costs.
Data Transfer Process - End to End
The Snow device workflow follows a well-defined process with cryptographic chain of custody at every step.
| Step | Action | Key Detail |
|---|---|---|
| 1. Order | Create a Snow job in the AWS console | Specify S3 bucket destination, KMS key, shipping address |
| 2. Receive device | AWS ships the device - typically 4-6 business days | Device arrives locked, unlocks with unlock code from console |
| 3. Connect and configure | Connect to local network, configure NFS or S3-compatible endpoint | Install AWS OpsHub or use Snowball client CLI |
| 4. Transfer data | Copy files to the device via NFS mount or AWS CLI | Data is encrypted with AES-256 using your KMS key |
| 5. Ship back | Drop off at UPS/DHL - E Ink shipping label auto-updates | Do not send with other packages - treat as high-value device |
| 6. Import | AWS receives device, imports data to your S3 bucket | Typically 1-2 days after AWS receives it |
| 7. Verification | Compare checksums in job completion report | AWS wipes the device securely after import |
# Configure Snowball client and copy data
# Install AWS OpsHub or use the Snowball Edge client
snowballEdge configure
# Enter: manifest file path, unlock code, IP address
snowballEdge unlock-device
# List NFS shares
snowballEdge describe-nfs-shares
# Mount NFS and copy data
sudo mount -t nfs -o nolock IP_ADDRESS:/SHARE_NAME /mnt/snowball
rsync -avz /source/data/ /mnt/snowball/
# Alternative: use AWS CLI with local endpoint
aws s3 cp /source/data/ s3://BUCKET_NAME/ \
--recursive \
--endpoint https://SNOWBALL_IP:8443 \
--ca-bundle certificate.pemDo not power off the Snowball device while a data transfer is in progress. Data in transit may not be committed. Always gracefully complete transfers before powering down.
Edge Computing with Snow Devices
Snow devices run a subset of AWS services locally using AWS Greengrass or EC2 AMIs, enabling compute at locations with limited connectivity.
| Capability | Snowcone | Snowball Edge Storage Optimized | Snowball Edge Compute Optimized |
|---|---|---|---|
| EC2 instances | Yes (SBE-C and SBE-S AMIs) | Yes (SBE-C and SBE-S AMIs) | Yes - more vCPU/RAM |
| AWS IoT Greengrass | Yes | Yes | Yes |
| Lambda (via Greengrass) | Yes | Yes | Yes |
| GPU inference | No | No | Optional NVIDIA V100 |
| Local S3-compatible API | Yes | Yes | Yes |
| Offline operation | Yes - no internet needed | Yes | Yes |
Snowball Edge Compute Optimized with GPU is used by customers who need to run machine learning inference models at locations without reliable internet - ships, mines, military forward operating bases, and manufacturing floors.
Snow Family Pricing and Transfer Economics
Snow pricing includes a service fee per job, shipping, and optionally on-site usage time. There is no charge for data uploaded to AWS via Snow (inbound transfer is free).
| Device | Service Fee | On-Site Usage | Notes |
|---|---|---|---|
| Snowcone | $60/job | $6/day after 3 days | Includes 3 days free on-site usage |
| Snowball Edge Storage Opt. | $300/job | $30/day after 10 days | Includes 10 days free on-site usage |
| Snowball Edge Compute Opt. | $300/job | $30/day after 10 days | GPU costs additional |
| Snowmobile | Custom pricing | Custom | Requires AWS account team engagement |
Data transfer economics - when does Snow beat the network?
| Data Size | 100 Mbps Link Time | 1 Gbps Link Time | Snow Time (approx) |
|---|---|---|---|
| 1 TB | 22 hours | 2 hours | 4-7 days (shipping + transfer) |
| 10 TB | 9 days | 22 hours | 4-7 days |
| 100 TB | 90 days | 9 days | 1-2 weeks (multiple devices) |
| 1 PB | 2.5 years | 90 days | 2-4 weeks (multiple Snowballs) |
Snow is typically cost-effective when you have more than 10 TB to transfer over a typical corporate internet connection. For ongoing large transfers, AWS DataSync over Direct Connect often makes more sense than repeated Snow shipments.
Interview Focus Points
- 1When would you recommend using Snow Family over AWS DataSync or Direct Connect for a data migration?
- 2Walk me through the end-to-end process of a Snowball Edge data transfer job.
- 3What are the differences between Snowcone, Snowball Edge Storage Optimized, and Snowball Edge Compute Optimized?
- 4How is data secured on a Snow device during transit?
- 5A client has 500 TB of on-premises data and a 100 Mbps internet connection - how would you plan the migration?
- 6What edge computing capabilities does Snowball Edge provide and what use cases does this enable?
- 7How would you handle a scenario where a Snowball device is lost in transit?