Q6: Migrating 70 TB of NFS Video Files to S3 With Minimal Bandwidth

Problem Info

Domain	Design Cost-Optimized Architectures
Services	AWS Snowball Edge, Amazon S3
Difficulty	Easy–Medium

Question

A company stores large video files (1 MB to 500 GB each) on on-premises NFS storage. Total data is 70 TB and no longer growing. The company wants to migrate everything to S3 as soon as possible, while using the least possible network bandwidth.

Which solution meets these requirements?

Correct Answer

Order an AWS Snowball Edge device. Receive it on-premises, use the Snowball Edge client to copy the data locally, then ship the device back so AWS can import the data into S3.

Why this is correct

Two requirements pull in what looks like opposite directions — “as soon as possible” and “least network bandwidth” — and Snowball Edge is the one AWS service built to satisfy both simultaneously for a large, one-time, finite dataset.

Because the 70 TB is static (not growing), this is a one-shot bulk migration, not an ongoing feed — exactly Snowball’s use case. Data transfer to the device happens over a local, high-speed connection (up to ~100 Gbps), so copying 70 TB takes roughly the low thousands of seconds — well under a couple of hours. Zero bytes cross the internet or corporate WAN link during that copy: bandwidth usage is effectively zero. The only “slow” part is physical shipping (a handful of business days each way), which is unavoidable for physically moving a device, but is still far faster and cheaper than saturating a corporate internet link for days or weeks trying to push 70 TB over the wire.

Why the alternatives fall short

• S3 Transfer Acceleration over the existing internet connection — great for ongoing, moderate-sized transfers (see Q1), but pushing 70 TB over any internet link — even an accelerated one — consumes significant bandwidth and could take a long time depending on available uplink speed, directly violating “least possible network bandwidth.”
• AWS DataSync — efficient for online, incremental, or ongoing NFS-to-S3 transfers, but it still moves data across the network. For a large one-time static dataset with a bandwidth constraint, it’s not the leanest choice compared to physically shipping a device.
• Set up Direct Connect — solves ongoing high-throughput, low-latency connectivity, but provisioning it takes weeks and is overkill (and costly) for a single 70 TB one-time migration.

Exam Tip

Pattern to memorize: large + one-time + limited/no bandwidth + on-premises → Snowball family. Scale further with data size:

• Snowcone — tens of terabytes, small/rugged form factor.
• Snowball Edge — up to ~80 TB usable per device (as in this question).
• Snowmobile — exabyte-scale, literally a shipping container on a truck.

If the dataset is actively growing / needs to sync continuously, that’s a signal for AWS DataSync instead, not Snowball.