LinkedIn

Cloud  /  AWS

AWS Amazon Web Services 61 guides · updated 2026

Hands-on guides to compute, storage, databases, networking, and serverless on the world's most widely adopted cloud platform.

AWS Snowball: Physical Data Transfer for When the Internet Is Not Fast Enough

At some point the math just does not work. You have 300 TB of data on-premises and a 1 Gbps internet connection. At 1 Gbps, with realistic overhead, transferring 300 TB takes about 27 days — and that assumes your entire internet connection is dedicated to the transfer with zero errors. In practice it takes months, often costs more in bandwidth fees than the migration budget allows, and leaves data exposed to network-level risks the whole time.

AWS Snowball exists because sometimes shipping a box is faster and cheaper than sending bits over the internet. That observation dates to 2015 when AWS launched the original Snowball device and it remains true today for datasets above roughly 10 TB.

The Snow Family

AWS groups its physical transfer devices under the Snow family, with different products for different scales:

AWS Snowcone: the smallest device in the lineup — 8 TB of usable storage (HDD version) or 14 TB (SSD version). Small enough to put in a backpack. Designed for edge locations where even larger Snowball devices would be hard to deploy: ships, oilfield sites, military forward operating bases, remote research stations. Can also transfer data over the network when connectivity is available, or be physically shipped to AWS.

AWS Snowball Edge Storage Optimized: 80 TB of usable block storage and S3-compatible object storage per device. Up to 10 devices can be clustered for a total of 800 TB. This is the workhorse for large-scale data migrations. Includes EC2 compute capability so you can run local pre-processing, encryption, or validation before the data reaches AWS.

AWS Snowball Edge Compute Optimized: less storage (28 TB) but more powerful compute — GPU option available. Designed for edge computing scenarios where you need significant processing power at a remote location: ML inference on surveillance footage, seismic data processing in the field, industrial analytics at a factory floor.

AWS Snowmobile: a 45-foot shipping container pulled by a semi-truck. Stores up to 100 PB. Used when a company is migrating an entire data center — exabyte-scale transfers that would take decades over the internet. AWS sends a truck to your location, fills the container, and drives it back.

  Snow Family — Capacity and Use Case
  =====================================


  Snowcone          Snowball Edge         Snowball Edge         Snowmobile
  (8-14 TB)         Storage Opt.          Compute Opt.          (100 PB)
                    (80 TB / device)      (28 TB)
  ┌───────┐         ┌─────────────┐       ┌─────────────┐       ┌─────────────────┐
  │ 📦    │         │ 📦📦📦📦   │       │ GPU ████    │       │   🚛 TRUCK      │
  │ Bag   │         │ Rack/pallet │       │ Local ML    │       │ 100 PB in       │
  │ carry │         │ up to 10    │       │ processing  │       │ a container     │
  └───────┘         └─────────────┘       └─────────────┘       └─────────────────┘
  Field/remote      Data migration        Edge AI/video         Datacenter
  locations         <800 TB               analysis              migration

When Physical Transfer Beats Network Transfer

The decision point is roughly:

  Transfer Time Comparison
  ========================


  Dataset: 100 TB


  Method                  | Time          | Cost estimate
  ------------------------|---------------|------------------
  100 Mbps internet       | ~90 days      | Bandwidth fees
  1 Gbps dedicated line   | ~9 days       | High line cost
  10 Gbps Direct Connect  | ~22 hours     | Expensive circuit
  Snowball Edge (1 device)| ~5 days total | ~$300 + storage
                          | (ship + load) |


  For 100+ TB, Snowball wins on both time and cost.
  Network transfer remains superior for <10 TB or ongoing sync.

The crossover point is roughly 10 TB on a standard internet connection. Above that, Snowball is typically faster and cheaper. Well below that, direct upload to S3 is simpler.

The Snowball Workflow

  1. Order: request a device through the AWS console. Specify the destination S3 bucket.
  2. AWS ships: a Snowball device arrives at your location, typically within days.
  3. Load data: connect the device to your network. Use the Snowball client or S3 adapter to copy data. The device encrypts everything with 256-bit AES using a KMS key.
  4. Ship back: use the prepaid shipping label to send the device to AWS.
  5. AWS imports: the data center technicians connect the device, verify the chain of custody, and import data to your S3 bucket. The device is then wiped using NIST 800-88 standards.

The data on the device is always encrypted. AWS never has your encryption key — it stays in KMS, associated with your account. If the device is intercepted in transit, the data is unreadable.

Edge Computing with Snowball

Beyond data transfer, Snowball Edge devices run local compute workloads via EC2-compatible instances and Lambda functions. This matters for locations without reliable internet connectivity:

  Edge Computing Pattern
  ======================


  Remote Site (no reliable internet)


  Sensors/Cameras ──► Snowball Edge ──► Local processing
                          │                (EC2 AMI, Lambda)
                          │
                     Local storage
                     (processed results)
                          │
                     When connectivity available:
                          │──► Upload results to S3
                     Or ship device to AWS:
                          └──► AWS imports full dataset

Real-World Use Case: Hospital Network Migration

A hospital group is migrating from an on-premises PACS (Picture Archiving and Communication System) to AWS. They have 2 PB of historical medical images — MRI scans, X-rays, CT scans — accumulated over 15 years. Regulatory requirements mandate that these images be preserved and accessible. Uploading 2 PB over the internet would take months and cost more in bandwidth than the project budget allows.

Solution: 25 Snowball Edge Storage Optimized devices. The hospital loads approximately 80 TB onto each device over one to two weeks (devices can be loaded in parallel), ships them back to AWS on a rolling basis. AWS imports them to S3, where a lifecycle rule moves images older than three years to Glacier for cost-effective long-term retention. Total migration time: six weeks instead of many months.

Key Interview Points