Snow Family — Concept
What it is
The AWS Snow Family is a set of physical devices AWS ships to your site for offline data transfer and edge compute in disconnected locations.
Why it exists
- Network transfer of TB to PB scales of data is slow and expensive over the internet.
- Some sites have no/low bandwidth (ships, mines, remote offices, military).
- You may need to process data at the edge before shipping or while disconnected.
Members
| Device | Storage | Compute | Use |
|---|---|---|---|
| Snowcone / Snowcone SSD | 8 / 14 TB usable | 4 vCPU, 4 GB RAM | Small/portable edge, IoT, drones, mobile field collection. Battery-friendly. |
| Snowball Edge — Storage Optimized | 80 TB usable | 40 vCPU, 80 GB RAM | Large data migration, periodic transfer |
| Snowball Edge — Compute Optimized | 28 TB | 52 vCPU, 208 GB RAM (+ optional GPU) | Edge ML / video / heavy compute disconnected |
| Snowmobile | up to 100 PB (one 45-ft truck) | — | Exabyte-class data center migration |
Snowmobile has been deprecated for new orders in some regions; Snowball Edge has effectively superseded it for most cases. Verify on exam day, but it still appears in SAA-C03 questions.
How a Snowball Edge job works
- Create a job in the AWS console → AWS ships the device.
- Connect on-prem network; Snowball client or OpsHub (GUI) uploads/downloads data.
- Encrypted with KMS end-to-end (256-bit, tamper-evident).
- Return prepaid shipping → AWS uploads to S3 (or pulls from S3 to ship out).
- Track via console + email.
Edge compute
- Snow devices can run EC2 AMIs, Lambda functions, and EKS Anywhere containers locally — useful for data preprocessing or running apps in disconnected sites.
When to use vs alternatives
| Need | Use |
|---|---|
| > 10 TB transfer, slow line | Snowball Edge |
| < 14 TB, very portable, battery | Snowcone |
| Petabyte-scale migration of a data center | Snowmobile (or many Snowballs in parallel) |
| Edge ML / IoT data preprocess | Snowball Edge Compute (with GPU option) |
| Continuous transfer over network | DataSync (or Direct Connect) |
| Permanent on-prem file share to AWS | Storage Gateway |
Rule of thumb (commonly tested)
Approx network days = Data size / Bandwidth. If shipping a Snowball (1-week turnaround) is faster than network transfer, use Snow. Example: 100 TB on a 1 Gbps line ≈ 10+ days saturated → Snowball wins (1 week round-trip + secure).
Common exam scenarios
- "Migrate 500 TB to S3, limited internet bandwidth" → Snowball Edge (multiple units in parallel).
- "Collect IoT data on a research vessel, intermittent connectivity" → Snowcone with on-board EC2/Lambda.
- "Decommission entire data center with 50 PB" → multiple Snowballs in parallel (or Snowmobile if available).
- "Run ML inference at a remote mine site" → Snowball Edge Compute with GPU.
- "Continuous nightly sync of 5 TB" → not Snow; use DataSync over DX/VPN.
Exam tip
"Petabytes and slow connection" = Snow. "Continuous / scheduled migration" = DataSync. "Permanent file share" = Storage Gateway.