Why not just upload over the network?
- Limited connectivity and bandwidth.
- High network costs.
- Shared bandwidth (cannot fully use the line).
- Connection instability.
Transfer Time Estimates
Data Size | 100 Mbps | 1 Gbps | 10 Gbps |
10 TB | 12 days | 30 hours | 3 hours |
100 TB | 124 days | 12 days | 30 hours |
1 PB | 3 years | 124 days | 12 days |
Exam Tip
If a network transfer takes more than ~1 week, use Snowball for offline migration.
Direct Upload vs Snowball Workflow
Direct Upload to S3
- Client → Amazon S3 bucket via internet.
- Assumes high-speed network (e.g., 10 Gbit/s).
- Limited by available bandwidth.
With Snowball
- Client writes data to Snowball device locally.
- Device is shipped to AWS.
- AWS imports/exports data into S3.
Benefit
Bypassing the internet removes bandwidth bottlenecks and stability issues.
Edge Computing with Snowball Edge
Definition
Process data on-site (where it’s generated) before sending to the cloud.
When to use
- Remote or mobile locations (ships, mining sites, moving vehicles).
- Limited/no internet connectivity.
- Need to process large volumes before upload.
How
- Deploy Snowball Edge Compute Optimized (or Storage Optimized with compute features).
- Run EC2 instances or AWS Lambda functions locally.
Use Cases
- Data preprocessing.
- Machine learning inference.
- Media transcoding.
Snowball → Glacier Architecture
Limitation
Snowball cannot write directly to Glacier.
Solution
- Import Snowball data into S3.
- Apply S3 Lifecycle Policy to transition data to Glacier or Glacier Deep Archive.
Data Flow
Snowball → Amazon S3 → (Lifecycle Policy) → Glacier